AI Distilled

98% cost reduction for GPT 4o miniAI_Distilled #70: OpenAI raises $6.6 billion funding, valuation at $157 billionThis 3 hour power packed workshop that will teach you 25+ AI Tools, make you a master of prompting & talk about hacks, strategies & secrets that only the top 1% know of.By the way, here’s sneak peek into what’s inside the workshop:-Making money using AI-The latest AI developments, like GPT o1-Creating an AI clone of yourself, that functions exactly like YOU-10 BRAND new AI tools to automate your work & cut work time by 50%Best thing? It's usually $399, but it's absolutely free for the first 100 readers.Save your seat now (Offer valid for 24 hours only)Welcome to AI_Distilled. Before we get to the newsletter, I have one quick message: Next week, we are hosting an AMA with Supreet Kaur: Navigating LLMs & AI Innovation. You should check it out.Today, we’ll talk about:Techwave:[Sponsored] Free 3 hour AI and ChatGPT workshop for professionalsOpenAI raises $6.6 billion funding, valuation at $157 billionOpenAI makes4 major announcements at DevDay, 98% cost reduction for GPT-4 to 4o miniMicrosoftlaunches redesigned Copilotwith Voice, Vision, and Chain of Thought capabilities.Metaunveils open-source Llama StackNotebookLM now summarizes YouTube videos. Andrej Karpathy'sNotebookLM tweet goes viralAwesome AI:Pika 1.5Graphite Code ReviewerHelicone:LLM-Observability for DevelopersMagic Patterns: Prototype your product ideas with AIRows: The new way to spreadsheetMasterclass:Anthropic reduces the error rate ofRAGs by 67% using this simple methodLangchain shows offnew tool: controllable Agentopen-source NotebookLM alternativeusing Llama 3.1 405BAndrew Ngannounces course on Meta's Llama 3.2, launching October 9Using task-specific models from AI21 Labs on AWSHackHub:o1-engineer: AI-powered code generation and editingCrawl4AI: LLM Friendly Web Crawler & ScraperLlama Stack:Model components of the Llama Stack APIsexo: Run your own AI cluster at home with everyday devicesTTS: a deep learning toolkit for Text-to-SpeechCheers!Shreyans SinghEditor-in-Chief, PacktLast Chance! For the next 48 hours only, save $150 on your full event pass!Use code LASTCHANCE40 at checkoutImagine being part of 10+ Power Talks, 12+ Hands-On Workshops, and 3 Interactive Roundtables—while networking with 30+ top industry leaders and hundreds of tech professionals from across the globe. This is your opportunity to dive into cutting-edge AI solutions at the Generative AI in Action 2024 Conference.It’s all happening November 11-13 (Virtual)—don’t miss your chance!BOOK YOUR SEAT NOW (before prices go up!)BOOK NOW AT $399.99 $239.99⚡ TechWave: AI/GPT News & AnalysisOpenAI raises $6.6 billion funding, valuation at $157 billionOpenAI has raised $6.6 billion in funding from investors like Microsoft, Nvidia, Thrive Capital, and Khosla Ventures, valuing the company at $157 billion. This significant investment comes as OpenAI restructures and undergoes leadership changes, including the departure of its CTO. Despite losses, OpenAI is projected to make $3.6 billion in revenue this year, with expectations for a major revenue increase next year. Investors are betting on the company's future growth, especially as it continues to pursue advanced AI goals like artificial general intelligence (AGI).OpenAI makes4 major announcements at DevDay, 98% cost reduction for GPT-4 to 4o miniAt OpenAI's 2024 DevDay, several key developer-focused features and tools were announced. One major update was prompt caching, offering a 50% discount on repeated prompts over 1,024 tokens, which lowers costs for developers automatically. Another significant launch was the WebSocket Realtime API, enabling real-time audio input/output for GPT-4 models, allowing developers to stream audio, text, and tool functions with low latency. OpenAI also simplified model distillation, making fine-tuning easier by allowing smaller models to learn from larger ones. Additionally, OpenAI extended free fine-tuning offers for GPT-4 models, and hinted at future support for image input through the Realtime API.Microsoftlaunches redesigned Copilotwith Voice, Vision, and Chain of Thought capabilities.Microsoft's October 2024 announcement highlights the evolution of Copilot. The updated Copilot integrates voice and vision capabilities, making interactions feel more natural and personalized. It offers practical help like summarizing news, taking notes at appointments, and assisting with life’s complexities. The tool aims to reduce information overload and provide a supportive, adaptive experience.Metaunveils open-source Llama StackMeta has introduced Llama Stack distributions to simplify the development of generative AI applications using its Llama large language models (LLMs). These distributions bundle multiple Llama Stack API providers into a single endpoint, allowing developers to work seamlessly with Llama models across different environments, including on-premises, cloud, and mobile devices. The Llama Stack provides essential building blocks for the entire AI development process, from model training to running AI agents.NotebookLM now summarizes YouTube videos. Andrej Karpathy'sNotebookLM tweet goes viralUsers can now upload videos or audio recordings, allowing NotebookLM to summarize key concepts and generate insights from these sources. It can transcribe and analyze audio or video content, creating helpful study guides or summaries. Additionally, users can now share Audio Overviews with a public link, making it easier to distribute content summaries.💻 Awesome AI: Tools for WorkPika 1.5Create stunning, cinematic video clips with advanced visual effects and longer scenes. It introduces new features like "Unreal Pikaffects," enabling users to manipulate objects in ways that go beyond real-life capture, such as exploding or inflating them. It also offers cinematic camera moves like Bullet Time and Crane Down, along with lifelike character actions like running or skateboarding.Graphite Code ReviewerGraphite Reviewer is an AI-powered tool that provides immediate, actionable feedback on pull requests, helping teams catch bugs, logical errors, and enforce best practices before human review. It integrates seamlessly with your codebase, offering code-aware suggestions without storing or using your team's data for training.Helicone / LLM-Observability for DevelopersHelicone is an open-source platform designed for developers to log, monitor, and debug large language models (LLMs). It provides tools for instant analytics, prompt management, and cost tracking, allowing users to filter, segment, and analyze their requests efficiently.Magic Patterns: Prototype your product ideas with AIMagic Patterns is an AI-powered design tool that allows users to quickly prototype product ideas by generating user interfaces (UIs) from prompts or images. It features an AI-native editor for iterating on components and designs, which can be exported to React or Figma.Rows — The new way to spreadsheetRows features an AI-powered assistant that helps users with tasks like data entry, classification, and translation, making it easier to work with complex information.🔛 Masterclass: AI/LLM TutorialsAnthropic reduces the error rate ofRAGs by 67% using this simple methodContextual Retrieval is an enhancement of traditional Retrieval-Augmented Generation (RAG) used in AI models to improve the accuracy of retrieving relevant information from large knowledge bases. Standard RAG uses embeddings to break down a knowledge base into chunks and retrieves relevant information based on semantic similarity. However, this method can lose important context, leading to retrieval errors. Contextual Retrieval addresses this by adding chunk-specific context before generating embeddings and BM25 (a ranking method based on exact matches), reducing retrieval errors by up to 67% when combined with reranking.Langchain shows offnew tool: controllable AgentThe Controllable-RAG-Agent is a sophisticated AI tool designed to answer complex questions using Retrieval-Augmented Generation (RAG) techniques. It employs a structured graph for reasoning and breaks down queries into smaller, manageable tasks. The agent ensures that answers are based solely on the provided data, preventing hallucinations, or incorrect content. It features multi-step reasoning, adapts its plan as new information is processed, and evaluates performance using metrics like answer correctness and relevance.open-source NotebookLM alternativeusing Llama 3.1 405BConvert your PDFs into podcasts with open-source AI models (Llama 3.1 405B, MeloTTS, Bark).Note: Only the text content of the PDFs will be processed. Images and tables are not included. The total content should be no more than 100,000 characters due to the context length of Llama 3.1 405B.Andrew Ngannounces course on Meta's Llama 3.2, launching October 9The course "Introducing Llama 3.2," offered by Amit Sangani, Senior Director of AI Partner Engineering at Meta, focuses on building multimodal applications using the Llama 3.2 family of models, which range from 1B to 405B parameters. It covers essential concepts from tokenization to tool-calling, as well as Llama's new stack, which simplifies application development.Using task-specific models from AI21 Labs on AWSIn this blog post, you'll learn how to use AI21 Labs' Task-Specific Models (TSMs) on AWS to streamline tasks like summarization, paraphrasing, and answering questions based on specific contexts. By subscribing to AI21 Labs in AWS Marketplace, setting up a SageMaker domain, and accessing these models through SageMaker JumpStart, you can easily deploy and customize them for your business. Unlike general foundation models, these TSMs are pre-trained for specific commercial tasks, offering greater accuracy and cost-efficiency with less need for complex prompt engineering.🚀 HackHub: AI Toolso1-engineer: AI-powered code generation and editingThe `o1-engineer` tool is a command-line utility that helps developers manage and interact with their projects more efficiently. It leverages OpenAI's API to automate tasks like code generation, file and folder management, project planning, and code review. By using commands like `/add`, `/edit`, and `/planning`, users can modify project structures, plan tasks, and streamline workflows directly from the terminal.Crawl4AI: LLM Friendly Web Crawler & ScraperCrawl4AI is an open-source, asynchronous web crawler designed to efficiently extract data for large language models (LLMs) and AI applications. It supports features like crawling multiple URLs simultaneously, extracting media and links, executing custom JavaScript, and managing sessions for dynamic web content. The tool allows for structured data extraction using CSS selectors or JSON strategies and offers advanced techniques for clustering and chunking content.Llama Stack:Model components of the Llama Stack APIsThe Llama Stack provides a set of APIs that cover the entire AI development lifecycle, including model training, inference, safety, memory management, and evaluation. Developers can mix and match local or cloud-based providers to implement these APIs, making it flexible for different use cases.exo: Run your own AI cluster at home with everyday devicesExo allows you to run AI models across multiple devices, like phones, laptops, or Raspberry Pis, forming a distributed AI cluster. It automatically discovers devices and splits model computations across them based on their resources. Unlike traditional systems with a master-worker architecture, Exo uses peer-to-peer connections, allowing all devices to contribute equally.TTS: a deep learning toolkit for Text-to-SpeechCoqui TTS is a deep learning toolkit for advanced text-to-speech (TTS) generation, designed for research and production use. It supports over 1,100 languages with pre-trained models and offers tools for training new models and fine-tuning existing ones. Coqui TTS includes various TTS models like Tacotron and Glow-TTS, speaker encoders for multi-speaker synthesis, and vocoders like MelGAN for high-quality audio output.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

xAI Colossus supercomputer with 100K H100 GPUs comes onlineAI_Distilled #66: OpenAI co-founder Sutskever's new safety-focused AI startup SSI raises $1 billion200+ hours of research on AI-led career growth strategies & hacks packed in 3 hoursThe only AI Crash Course you need to master 20+ AI tools, multiple hacks & prompting techniques in just 3 hoursYou’ll save 16 hours every week & find remote jobs using AI that will pay you upto $10,000/moGet It Here For Free (Valid For Next 24 hours Only!)Welcome to AI_Distilled. Today, we’ll talk about:Techwave:[Sponsored] 3-hour Mini Course on AI (worth $399) for FREEOpenAI co-founder Sutskever's new safety-focused AI startup SSI raises $1 billionxAI Colossus supercomputer with 100K H100 GPUs comes onlineOpenAI Japan announces next-generation model 'GPT Next'100M Token Context Windows is here350M downloads of Llama since 2023Awesome AI:Build web applications quickly by generating front-end codePowerful APIs for speech-to-text, text-to-speech, and language understandingv0 by VercelRevolutionize Your Storyboarding ProcessMeasure developer shipping velocity, accuratelyMasterclass:Natural Language Processing and Machine Learning for DevelopersBuild a generative AI image description applicationVisualizing and interpreting decision treesRethinking the Role of PPO in RLHFEnhancing Paragraph Generation with a Latent Language Diffusion Model Transparency is often lacking in datasets used to train large language modelsHackHub:A natural language interface for computersLLM app development platform2^x Image Super-ResolutionVideo generation platform based on diffusion modelsPop Audio-based Piano Cover GenerationCheers!Shreyans SinghEditor-in-Chief, PacktLive Webinar: The Power of Data Storytelling in Driving Business Decisions (September 10, 2024 at 9 AM CST)Data doesn’t have to be overwhelming. Join our webinar to learn about Data Storytelling and turn complex information into actionable insights for faster decision-making.Click below to check the schedule in your time zone and secure your spot. Can't make it? Register to get the recording instead.REGISTER FOR FREE⚡ TechWave: AI/GPT News & AnalysisOpenAI co-founder Sutskever's new safety-focused AI startup SSI raises $1 billionSafe Superintelligence (SSI), co-founded by Ilya Sutskever, who was previously the chief scientist at OpenAI. SSI has raised $1 billion in funding to develop safe AI systems that surpass human abilities. The company, valued at $5 billion, plans to use the money for computing power and hiring top talent. Sutskever, along with Daniel Gross and Daniel Levy, started SSI in June 2024.xAI Colossus supercomputer with 100K H100 GPUs comes onlineElon Musk's X (formerly Twitter) has brought online the world's most powerful AI training system, called Colossus, using 100,000 Nvidia H100 GPUs. The supercomputer will soon expand with an additional 50,000 H100 and H200 GPUs, bringing the total to 200,000. Developed by Dell in just 122 days, Colossus will be used for training advanced AI models, such as xAI's Grok version 2.OpenAI Japan announces next-generation model 'GPT Next'Tadao Nagasaki, CEO of OpenAI Japan, announced that ChatGPT has reached over 200 million active users by the end of August, marking it as the fastest software in history to reach this milestone. He highlighted the growing adoption of ChatGPT Enterprise among companies like Apple, Coca-Cola, and Moderna. Nagasaki also discussed OpenAI's future plans, introducing the next-generation AI model, "GPT Next," which he claims will be 100 times more powerful than previous models like GPT-4, supporting advanced capabilities across various data formats.100M Token Context Windows is hereMagic has developed ultra-long context AI models, capable of processing up to 100 million tokens of context during inference, which could revolutionize tasks like code synthesis. To improve testing, Magic introduced HashHop, a method that eliminates these oversights by using random hashes, forcing models to store and retrieve complex information. Magic also announced new partnerships with Google Cloud and NVIDIA to scale AI infrastructure and raised $465M to support their work.350M downloads of Llama since 2023Meta's Llama models have rapidly become one of the most widely used open-source AI model families, with over 350 million downloads, driven by its availability on platforms like Hugging Face and partnerships with major cloud providers like AWS and Azure. Llama 3.1 has expanded its capabilities, offering enhanced context lengths, multilingual support, and new safety tools. Its open-source nature encourages innovation, with companies like AT&T, DoorDash, and Accenture using Llama to enhance customer experiences, streamline operations, and drive AI-powered solutions across industries.💻 Awesome AI: Tools for WorkGPT EngineerBuild web applications quickly by generating front-end code using technologies like React, Tailwind, and Vite. Users can describe their app ideas, sync them with GitHub, and deploy them with a single click.OpenHomeAI-powered voice interface that enables natural, seamless conversations with devices using its Voice SDK, allowing any platform to integrate smart voice control. It offers powerful APIs for speech-to-text, text-to-speech, and language understanding, making it ideal for applications like medical transcription and smart home automation. 500 features, including instant translation, emotion detection, and media control.v0 by VercelGenerate web development components and full interfaces quickly using chat-based prompts. It helps developers create UI elements like buttons, modals, and pages by simply describing what they need, enabling faster development workflows.StoryboarderRapidly transform ideas into detailed storyboards, animatics, and screenplays. With features like Image-To-Video, the platform can turn static images into dynamic videos, enhancing storytelling and saving time. It supports various media projects, including commercials, films, and social media content, and offers integrated scriptwriting, consistent art styles, and expert support to streamline the creative process.Maxium AIAccurately measure developer efficiency by tracking shipping velocity and performance, going beyond just lines of code or commits. It integrates with GitHub to provide a standardized evaluation mechanism across different tech stacks and programming languages.🔛 Masterclass: AI/LLM TutorialsBuild a generative AI image description applicationThis guide explains how to build an application for generating image descriptions using Anthropic's Claude 3.5 Sonnet model on Amazon Bedrock and AWS CDK. By integrating Amazon Bedrock’s multimodal models with AWS services like Lambda, AppSync, and Step Functions, you can quickly develop a solution that processes images and generates descriptions in multiple languages. The use of Generative AI CDK Constructs streamlines infrastructure setup, making it easier to deploy and manage the application.Visualizing and interpreting decision treesTensorFlow recently introduced a tutorial on using dtreeviz, a leading visualization tool, to help users visualize and interpret decision trees. dtreeviz shows how decision nodes split features and how training data is distributed across different leaves. For example, a decision tree might use features like the number of legs and eyes to classify animals. By visualizing the tree with dtreeviz, you can see how each feature influences the model's predictions and understand why a particular decision was made.Rethinking the Role of PPO in RLHFIn Reinforcement Learning with Human Feedback (RLHF), there's a challenge where the reward model uses comparative feedback (i.e., comparing multiple responses) while the fine-tuning phase of RL uses absolute rewards (i.e., evaluating responses individually). This discrepancy can lead to issues in training. To address this, researchers introduced Pairwise Proximal Policy Optimization (P3O), a new method that integrates comparative feedback throughout the RL process. By using a pairwise policy gradient, P3O aligns the reward modeling and fine-tuning stages, improving the consistency and effectiveness of training. This approach has shown better performance in terms of reward and alignment with human preferences compared to previous methods.Enhancing Paragraph Generation with a Latent Language Diffusion Model The PLANNER model, introduced in 2023, enhances paragraph generation by combining latent semantic diffusion with autoregressive techniques. Traditional models like GPT often produce repetitive or low-quality text due to "exposure bias," where the training and inference processes differ. PLANNER addresses this by using a latent diffusion approach that refines text iteratively, improving coherence and diversity. It encodes paragraphs into latent codes, processes them through a diffusion model, and then decodes them into high-quality text. This method reduces repetition and enhances text quality.Transparency is often lacking in datasets used to train large language modelsA recent study highlights the lack of transparency in datasets used to train large language models (LLMs). As these datasets are combined from various sources, crucial information about their origins and usage restrictions often gets lost. This issue not only raises legal and ethical concerns but can also impact model performance by introducing biases or errors if the data is miscategorized. To address this, researchers developed the Data Provenance Explorer, a tool that provides clear summaries of a dataset’s origins, licenses, and usage rights.🚀 HackHub: AI ToolsOpenInterpreter/open-interpreterOpen Interpreter is a tool that allows language models (like GPT-4) to execute code locally on your machine, supporting languages like Python, JavaScript, and shell scripts. It works like ChatGPT but with the ability to interact with your system's resources.langgenius/difyDify is an open-source platform for developing AI applications using large language models (LLMs). It provides an intuitive interface for building AI workflows, managing models, and integrating tools like Google Search or DALL·E. Dify supports a wide variety of LLMs and offers features like a prompt IDE, document retrieval (RAG), agent-based automation, and detailed observability for monitoring performance.Tohrusky/Final2xFinal2x is a cross-platform tool designed to enhance image resolution and quality using advanced super-resolution models such as RealCUGAN, RealESRGAN, and Waifu2x. It's ideal for anyone looking to improve image resolution efficiently across various platforms.ali-vilab/VGenVGen is an open-source video generation platform from Alibaba's Tongyi Lab that offers a wide range of tools for generating videos from various inputs like text, images, and motion instructions. It features state-of-the-art models like I2VGen-xl for image-to-video synthesis and DreamVideo for custom subject and motion generation. VGen supports tasks like video generation from human feedback and video latent consistency modeling.sweetcocoa/pop2pianoPop2Piano is a deep learning model that automatically generates piano covers from pop music audio. Traditionally, creating a piano cover involves understanding the song's melody, chords, and mood, which is challenging even for humans. Prior methods used melody and chord extraction, but Pop2Piano skips these steps, directly converting pop music waveforms into piano covers using a Transformer-based approach. The model was trained on a large dataset of synchronized pop songs and piano covers (300 hours), enabling it to generate plausible piano performances without explicit musical extraction modules.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Miss this week’s AI news and you might just fall behind.AI_Distilled #96: What’s New in AI This WeekYou can now run and fine-tune Qwen3 and Meta's new Llama 4 models with 128K context length & superior accuracy. Unsloth is an open-source project that allows easy fine-tuning of LLMs and that also uploads accurately quantized models to Hugging Face. GitHub repo: https://github.com/unslothai/unslothUnsloth's new Dynamic 2.0 quants outperform other quantization methods on 5-shot MMLU & KL Divergence benchmarks, meaning you can now run + fine-tune quantized LLMs while preserving as much precision as possible. Read more here . Tutorial for running Qwen3 here.Tutorial for running Llama 4 here.Welcome to another exciting edition of our AI_Distilled! This week, we're witnessing a surge in innovative AI solutions, with companies like OpenAI and Microsoft rolling out tools that streamline development and enhance user interaction. From Apple opening its models to developers to the fierce competition for AI's top talent, join us as we explore the latest breakthroughs shaping our digital world.LLM Expert Insights,PacktIn today's issue:📅 June’s AI Must-Attends: From AI Engineer World’s Fair to Packt’s Agent Bootcamp—here are 6 events you don’t want to miss this month.🔌 MCP, Explained: Paul Singh breaks down the Model Context Protocol—your plug-and-play solution for seamless AI tool integration.💻 Codex Arrives: OpenAI rolls out Codex, a powerful AI coding agent for writing features, fixing bugs, and navigating codebases.🧠 Windows Gets Smarter: Microsoft integrates native MCP into Windows and launches AI Foundry for seamless agent automation.🎟️ Google AI Ultra Drops: A new $249.99/mo subscription offers Gemini upgrades, cinematic video tools, and 30TB of storage.🍏 Apple Opens Up: Developers may soon build apps with Apple’s AI models—announcement expected at WWDC 2025.🏁 AI Talent Wars: OpenAI, Google & more compete for elite researchers—offering private jets and millions in perks.👨‍💻 Copilot’s New AI Agent: GitHub's upgraded Copilot now tackles coding issues with draft PRs, vision models, and full MCP support.🎧 On-Device Audio AI: Stability AI & Arm launch a mobile-ready model for text-to-audio generation—11 seconds of sound in 8.📈EXPERT INSIGHTSJUNE'S MUST ATTEND AI/LLM EVENTSIn June 2025, a number of exciting AI conferences are already generating buzz. Here are the Top 5 not-to-miss events in the next month (for more information and registration details, please visit the links):1. AI Engineer World’s FairDate: June 3–5, 2025Location: San Francisco, California, USACost: $299–1,799 in-personThe AI Engineer World's Fair, from June 3-5, 2025, in San Francisco, is the largest technical conference for AI engineers. It would host approximately 3,000 attendees, featuring 150 talks and 100 practical workshops. Topics include Generative AI, AI agents, LLMs, infrastructure, and AI in Fortune 500 companies, offering unparalleled networking and learning opportunities for industry professionals.2. Data + AI SummitDate: June 9–12, 2025Location (Hybrid): San Francisco, California, US, and available online.Cost: $1,395–1,895 in-person. Free for virtual admission. Discounted tickets are available with group-rate pricing.The Data + AI Summit is a four-day event hosted by Databricks. It includes panel discussions, networking opportunities, and training workshops on topics such as data engineering, data governance, and machine learning.3. The AI Summit LondonDate: June 11–12, 2025Location: Tobacco Dock, London, UKCost: £125–2,499AI Summit London, spanning over two days, will cover a wide range of topics including agentic AI in action and ethical use of AI. With a strong lineup of sponsors and thousands of guests, the summit offers great opportunities for networking with leading AI practitioners.4. Packt’s AI Agent Bootcamp (Build AI Agents Over the Weekend)Date: June 21–22 and 28–29, 2025Location: Live Virtual WorkshopCost: Our AI Agent Bootcamp aims to equip developers, ML engineers, data scientists, technical professionals, and software architects with the practical skills to design, build, and deploy AI agents using frameworks like LangChain, AutoGen, and CrewAI, moving from theoretical understanding of LLMs to practical application.5. CDAO GovernmentDate: June 25–26, 2025Location: Washington, D.C., USCost: $499 in-person; Free for VP and C-level government executives.The CDAO Government conference in Washington, D.C., is unique as it unites U.S. government data leaders to explore AI, governance, and ethical data use in public services. Celebrating its 13th anniversary, this event offers an excellent opportunity to learn how to securely leverage AI's capabilities for government data challenges.This was just a quick peek into spaCy pipelines — but there’s much more to explore.For instance, the spacy-transformers extension integrates pretrained transformer models directly into your spaCy pipelines, enabling state-of-the-art performance. Additionally, the spacy-llm plugin allows you to incorporate LLMs like GPT, Cohere, etc. for inference and prompt-based NLP tasks.Master AI Tools, Set Automations & Build Agents – all in 16 hours (for free)AI is no longer just a buzzword — it’s the most valuable skill of this decade– to make money, to get hired and to be future-paced.That’s why, you need to join the 2-Day Free AI Upskilling Sprint by Outskill which comes with 16 hours of intensive training on AI frameworks, tools and tactics that will make you an AI expert.Originally priced at $499, but the first 100 of you get in for completely FREE! Claim your spot now for $0! 🎁📅23rd May- Kick Off Call & Session 1✅Live sessions- 24th & 25th May🕜11AM EST to 7PM ESTJOIN NOW(Limited Free Seats! 🚨)EXPERT INSIGHTS BY PAUL SINGHModel Context Protocol (MCP) and what it means for youIf you're working on AI design or tool integration, the Model Context Protocol (MCP) offers a seamless, standardized way to connect AI tools, data sources, and LLM applications. Developed by Anthropic, MCP is an open protocol designed to simplify the often complex and time-consuming process of integrating rapidly evolving AI models with tools and services. Think of it as the USB-C of the AI world—plug-and-play, regardless of the LLMs or tools you're working with, and without diving into the intricate technicalities of MCP itself.MCP operates on a client-server model, where your LLM application runs a local MCP client that communicates with one or more MCP servers. A service provider only needs to implement a single MCP server, which can then handle APIs, databases, and other services, without requiring constant code adjustments for each new integration.Take a look at how three different MCP servers integrate with APIs and services:MCP leverages the lightweight JSON-RPC message format (a simple remote procedure call protocol), stateful connections, server-client capability negotiation, and reflection. Reflection allows the client to query the server about its capabilities, which can then be surfaced to the LLM automatically via the orchestrating application’s prompt.When designing with MCP, it's important to keep your architecture modular, test each component thoroughly, document your iterations, and ensure security by validating inputs and controlling access.MCP is gaining traction with large organizations like Microsoft, which is integrating it into key products such as Semantic Kernel, Copilot Studio, and GitHub Copilot. I envision a near future where MCP-as-a-Service becomes the de facto standard, eliminating deployment overhead and enabling seamless AI-to-AI or agent-to-agent communication. For example, MCP endpoints could allow straightforward integration without server management, while internal repositories of MCP clients could democratize standardized tool access across organizations.To read more about MCP, you can check out these resources: https://modelcontextprotocol.io and https://aka.ms/mcp. I’ll continue to share how our customers and various industries are adopting MCP and the lessons we’re learning along the way. Stay tuned for more.Join Packt’s Accelerated Agentic AI Bootcamp this June and learn to design, build, and deploy autonomous agents using LangChain, AutoGen, and CrewAI. Hands-on training, expert guidance, and a portfolio-worthy project—delivered live, fast, and with purpose.This is it.35% off this Workshop - Limited Time OfferIf you’re in—move now.Code: AGENT35RESERVE YOUR SEAT NOW!📈LATEST DEVELOPMENTOpenAI Introduces Codex for Enhanced Code GenerationOpenAI has released Codex, a cloud-based AI agent for software engineering. Available in ChatGPT Pro, Enterprise, and Team, Codex (powered by codex-1) can write features, fix bugs, and answer codebase questions, operating in isolated environments. It learns from real-world tasks, producing human-like code and iteratively running tests. Developers can monitor progress, review changes with verifiable evidence, and guide Codex with AGENTS.md files.Microsoft Unveils Windows AI Foundry and Native MCP for Future AI AgentsMicrosoft is advancing its AI vision with native Model Context Protocol (MCP) in Windows and the Windows AI Foundry. This crucial groundwork, leveraging Anthropic's "USB-C of AI" protocol, aims to enable automated AI agents to seamlessly interact with apps, web services, and Windows functions. This initiative will empower features like natural language file searches and AI-powered system controls, reshaping how users engage with their devices.Google Launches AI Ultra: A VIP Pass to Advanced AIGoogle is launching Google AI Ultra, a new $249.99/month subscription (with an initial discount) offering the highest usage limits and access to its most capable AI models and premium features. Tailored for creative professionals, developers, and researchers, it includes Gemini with enhanced reasoning, Flow for cinematic video creation, Whisk for animated image generation, and advanced NotebookLM. Subscribers also get Gemini integration in Google apps (Gmail, Docs, Chrome), Project Mariner for multi-task management, YouTube Premium, and 30 TB storage.Apple to Open AI Models for DevelopersApple is reportedly preparing to allow third-party developers to build software using its AI models, aiming to boost new application creation. This move, expected to be unveiled at WWDC on June 9th, would let developers integrate Apple's underlying AI technology into their apps, starting with on-device models. This could help Apple compete in the AI landscape and enhance Apple Intelligence's appeal.GitHub Copilot Launches New AI Coding AgentGitHub Copilot now features an AI coding agent that tackles low-to-medium complexity tasks by simply assigning it issues. It operates in secure, customizable environments, pushing commits to draft pull requests with transparent session logs. This agent, enhanced by Model Context Protocol (MCP) and vision models, allows developers to offload routine work, ensuring security through human approval for pull requests and adhering to existing policies.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️We would love to know what you thought—your feedback helps us keep leveling up.👉 Drop your rating hereThanks for reading,The AI_Distilled Team(Curated by humans. Powered by curiosity.)*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0}#converted-body .list_block ol,#converted-body .list_block ul,.body [class~=x_list_block] ol,.body [class~=x_list_block] ul,u+.body .list_block ol,u+.body .list_block ul{padding-left:20px} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Replit Agent early accessAI_Distilled #67: Apple Intelligence comes to iPhone, iPad, and Mac starting next monthGrow your business & career by 10x using AI Strategies in 4 hrs! 🤯Imagine a future where your business runs like a well-oiled machine, effortlessly growing and thriving while you focus on what truly matters.This isn't a dream—it's the power of AI, and it's within your reach.Join our AI Business Growth & Strategy Crash Course and discover how to revolutionize your approach to business on 12th September at 10 AM EST.In just 4 hours, you’ll gain the tools, insights, and strategies to not just survive, but dominate your market.Sign up here to save your seat! 👈Welcome to AI_Distilled. Today, we’ll talk about:Techwave:[Sponsored] Grow your career by 10x using AI Strategies in 4 hrs!Apple Intelligence comes to iPhone, iPad, and Mac starting next monthReplit Agent early accessAI system developed by Google DeepMind that designs novel proteinsIntroducing LLaVA V1.5 7B on GroqCloudFunction Calling in Google AI StudioAwesome AI:Polymet - Idea to prototype within secondsClipAnything - Choppityfal.aiEarkick - Your Personal AI ChatbotOuterbase | The interface for your databaseMasterclass:Voice Trigger System for SiriAlign Meta Llama 3 to human preferences with DPOAn Intuitive Intro to RLEnhancing LLMs with Structured Outputs and Function CallingSafely repairing broken builds with MLHackHub:Agents for software development Open-source LLM app development platformbuild, manage & run useful autonomous agentsUnderstand Human Behavior to Align True NeedsGenerative models for conditional audio generationCheers!Shreyans SinghEditor-in-Chief, Packt💡Recommended Reading: Essential Concepts of Vector DatabasesUnderstand why vector databases are important in modern data management and how to use them effectively.The course is about 4 hours long and is aimed at people interested in advanced data management techniques.The course includes hands-on sessions for setting up and using these databases, as well as integrating them with Large Language Models and frameworks like LangChain.Get it for $84.99⚡ TechWave: AI/GPT News & AnalysisApple Intelligence comes to iPhone, iPad, and Mac starting next monthApple announced the launch of "Apple Intelligence," a personal intelligence system integrated with iOS 18, iPadOS 18, and macOS Sequoia, starting in October 2024. This system uses advanced generative models and personal context to enhance everyday tasks, like writing assistance, smarter notifications, and a more flexible Siri. Features like a photo Clean Up tool, transcription in Notes and Phone apps, and AI-powered email prioritization will debut first in the U.S., with expanded language and feature support in the following months.Replit Agent early accessReplit Agent is an AI tool that helps users create software projects by understanding natural language prompts. Currently in early access for Replit Core and Teams subscribers, it assists in building web-based applications by guiding users through each step, from selecting technologies to deploying the final product. The agent is designed for prototyping and works closely with users to refine and develop their applications.AI system developed by Google DeepMind that designs novel proteinsAlphaProteo is an AI system developed by Google DeepMind that designs novel proteins to bind to specific target molecules. This technology can accelerate biological research by creating protein binders that aid in drug development, disease understanding, and more. AlphaProteo builds on the success of AlphaFold but goes further by generating new proteins, not just predicting their structures. It has shown high success rates in binding to key targets, such as proteins involved in cancer and viral infections like SARS-CoV-2.Introducing LLaVA V1.5 7B on GroqCloudLLaVA v1.5 7B is a new multimodal AI model available on GroqCloud, enabling developers and businesses to create applications that integrate image, audio, and text inputs. Built from a combination of OpenAI’s CLIP and Meta’s Llama 2, LLaVA v1.5 excels in tasks like visual question answering, image captioning, and multimodal dialogue.Function Calling in Google AI StudioGoogle AI Studio now supports function calling, allowing users to easily test the model's capabilities directly in the interface. This new feature makes it more convenient to experiment with the AI without leaving the UI. Google AI Studio offers free fine-tuning.💻 Awesome AI: Tools for WorkPolymet - Idea to prototype within secondsPolymet is an AI-powered tool that helps users quickly turn ideas into prototypes by generating designs and production-ready code in seconds. Users can describe what they need, iterate on the design with their team, and then export the code and designs, which can easily integrate with tools like Figma and existing codebases.ClipAnything - ChoppityChoppity is an AI-powered video editing tool that allows users to quickly find and clip moments from any video using visual, audio, and sentiment analysis. With its "ClipAnything" feature, users can search for specific parts of a video, such as key events, people, or emotions, without having to manually review hours of footage.fal.aiFal.ai is a generative media platform designed for developers to create and deploy AI-powered applications, particularly focused on text-to-image models. It offers fast, cost-effective inference with models like FLUX.1 and Stable Diffusion, optimized for various creative tasks.Earkick - Your Personal AI ChatbotEarkick is an AI-powered mental health app that helps users track and improve their emotional well-being in real time through a personal chatbot named Panda. Earkick tracks mental readiness, mood, and calmness, while providing daily insights, breathing techniques, and guided self-care sessions.Outerbase | The interface for your databaseOuterbase is an AI-powered platform that simplifies working with databases for engineers, researchers, and analysts. It supports SQL and NoSQL databases, allowing users to manage data securely while using AI tools to write queries, fix mistakes, and generate charts and visualizations instantly. Outerbase's table editor, dashboards, and data catalog help users organize, analyze, and share insights efficiently.🔛 Masterclass: AI/LLM TutorialsVoice Trigger System for SiriApple's voice trigger system for Siri includes a first-stage low-power detector to identify potential triggers, and a second-stage, high-precision model to confirm the trigger. It also incorporates speaker identification to ensure the device responds only to its primary user. This sophisticated setup addresses challenges like background noise and phonetically similar words while maintaining power efficiency and privacy.Align Meta Llama 3 to human preferences with DPODPO involves fine-tuning a large language model (LLM) based on feedback from human annotators who rate or rank the model's responses according to desired values, such as helpfulness and honesty. SageMaker Studio provides the computational environment to fine-tune the model using Jupyter notebooks with powerful GPU instances, while SageMaker Ground Truth simplifies the process of gathering human feedback by managing workflows for data annotation. Together, they allow you to align the Llama 3 model’s responses with specific organizational values efficiently.An Intuitive Intro to RLReinforcement learning (RL) is a type of machine learning where an agent learns by interacting with its environment, making decisions, and receiving feedback in the form of rewards or penalties. The goal is to maximize cumulative rewards over time. The agent starts with little to no knowledge and improves through trial and error, learning from past experiences. In RL, actions taken by the agent change the state of the environment, and based on the rewards received, the agent adjusts its future actions. A key concept in RL is balancing exploration (trying new things) and exploitation (using known strategies for rewards).Enhancing LLMs with Structured Outputs and Function CallingEnhancing LLMs with structured outputs and function calling improves their ability to provide accurate and useful responses. Structured outputs ensure consistency and clarity by organizing information in a logical format, reducing ambiguity. Function calling allows LLMs to perform specific tasks, such as retrieving real-time data or executing external functions, making them more interactive and versatile. Combined with techniques like Retrieval-Augmented Generation (RAG), which integrates relevant external information into the model’s responses, these enhancements lead to more reliable, accurate, and contextually rich conversations with LLMs.Safely repairing broken builds with MLGoogle's engineers have developed a machine learning model called DIDACT to automatically repair broken code builds by analyzing historical data of build errors and their fixes. This model suggests potential fixes to developers directly within their Integrated Development Environment (IDE). In a controlled experiment, the use of these machine learning-suggested fixes improved productivity by reducing active coding and feedback time, and increasing the number of completed code changes.🚀 HackHub: AI ToolsAll-Hands-AI/OpenHandsOpenHands is an AI-powered platform designed to assist with software development, allowing agents to perform tasks similar to human developers. These agents can modify code, run commands, browse the web, call APIs, and even use resources like StackOverflow. OpenHands is easy to set up using Docker and can be run in various modes, including scriptable or interactive CLI.langgenius/difyDify is an open-source platform for developing AI applications, offering an intuitive interface that integrates workflows, agent capabilities, model management, and observability features. Dify's core features include a visual AI workflow builder, integration with numerous LLMs, agent tools, and a retrieval-augmented generation (RAG) pipeline for document handling.TransformerOptimus/SuperAGISuperAGI is an open-source framework designed for developers to create, manage, and run autonomous AI agents. It allows seamless operation of multiple agents simultaneously and provides tools to extend their capabilities. With features like graphical interfaces, performance telemetry, and integration with multiple vector databases, SuperAGI enables AI agents to efficiently handle tasks, learn from experience, and optimize token usage.lllyasviel/Paints-UNDOPaints-Undo is an open-source project that provides AI models designed to simulate the drawing process in digital art. By inputting a completed image, users can generate a sequence of steps showing how that image might have been created, mimicking the "undo" function in digital painting software.Stability-AI/stable-audio-toolsStable-Audio-Tools is an open-source library for working with audio generation models. It provides tools for training and running models that generate audio, including a Gradio interface for testing. Users can install the library via PyPI, and the repository includes scripts for both training models and performing inference.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

How to boost LLM performance during pre-training: A preview AI_Distilled #97: What’s New in AI This Week Build Your AI Chatbot with Free LLM Boomcamp Join LLM Zoomcamp, a free online course starting on June 2 and build an end-to-end AI chatbot tailored to your use case. In 10 weeks, you’ll learn key skills like working with LLMs and RAG, vector search for indexing and retrieval, how to evaluate and monitor performance, and key best practices for building robust, real-world applications. REGISTER NOW FOR FREE It’s time for the final issue of May 2025. In this edition, we bring you the top five news highlights of the week, upcoming events shaping the AI and LLM landscape, and a sneak peek into techniques for optimizing LLM performance. LLM Expert Insights, Packt In today's issue: 🧠 Expert Deep Dive: This week, we explore pre-training optimization techniques—from quantization to flash attention—for building faster, smarter LLMs. 📅 Webinar Watchlist: June’s top AI/LLM webinars cover automation, cybersecurity, healthcare, legal AI, and multimodal fine-tuning. 🔌 Build AI Agents This Weekend: Join Packt’s Accelerated Agentic AI Bootcamp—hands-on, fast-paced, and 35% off. 📚 Optimize Your LLM Stack: Learn more from Generative AI with Python and PyTorch—a guide to efficient training and deployment. 🚀 DeepSeek V3 Debuts: China’s latest open-source model steps up with better reasoning and dev capabilities. 📰 Publishers vs. AI Search: Google CEO Sundar Pichai defends AI-powered results amid growing backlash from content creators. 📱 Apple Rebrands for 2026: WWDC will unveil iOS 26 and align all platforms under a unified OS naming strategy. 🎨 Sam Altman x Jony Ive: OpenAI teams up with the design legend to build magical, AI-first consumer products. 🧠 Anthropic Traces Thoughts: Claude’s internal reasoning gets visualized through groundbreaking interpretability research. 📈UPCOMING EVENTS JUNE'S MUST ATTEND AI/LLM WEBINARS In June 2025, a number of exciting AI webinars are already generating buzz. Here are the Top 5 not-to-miss events in the next month (for more information and registration details, please visit the links): 1. AI-Enhanced Motion Control: Innovations Driving Automation Forward Date: June 5, 2025 Time: 12:00 PM – 1:00 PM ET Location: Online Cost: Free Hosted by the Association for Advancing Automation, this webinar explores how AI is revolutionizing motion control systems, enhancing precision, efficiency, and adaptability across various industries. 2. AI Security Webinar – Practical Measures to Mitigate AI and Cybersecurity Risks Date: June 11, 2025 Time: 11:00 AM – 12:30 PM BST Location: Online Cost: Free Presented by The Alan Turing Institute, this interactive webinar brings together industry experts and SMEs to share practical, cost-efficient, and high-impact security measures that deliver maximum AI and cybersecurity protection for businesses. 3. Clinical Large Language Models in Healthcare – Applications, Challenges, and Opportunities Date: June 12, 2025 Time: 10:00 AM – 11:00 AM CEST Location: Online Cost: Free Organized by the Helmholtz Information & Data Science Academy in collaboration with NORA, this webinar features Anne Torill Nordsletta discussing the role of large language models in healthcare, exploring applications, challenges, and future opportunities in the clinical setting. 4. Inside the TBI Playbook: How I Use AI to Win the Hardest Cases Date: June 17, 2025 Time: 1:00 PM – 2:30 PM EST Location: Online Cost: Free Hosted by Anytime AI™, this CLE-accredited webinar features attorney Taylor Ernst sharing insights on leveraging AI in traumatic brain injury litigation. Attendees will learn about practical applications of AI tools in complex legal cases. 5. Multi-Modal LLM Fine-Tuning of Unstructured Data with Dataloop & SingleStore Date: June 18, 2025 Time: 10:00 AM – 11:00 AM PST Location: Online Cost: Free Presented by SingleStore, this webinar explores techniques for fine-tuning multi-modal large language models on unstructured data, covering integration strategies with Dataloop and SingleStore platforms. Machine Learning Summit 2025 JULY 16–18 | LIVE (VIRTUAL) 20+ ML Experts | 20+ Sessions | 3 Days of Practical Machine Learning and 35% OFF BOOK NOW AND SAVE 35% Use Code EMAIL35 at checkout when purchasing the 3-day ticket Limited to the first 50 customers EXPERT INSIGHTS PRE-TRAINING OPTIMIZATION TECHNIQUES FOR LLMs The scale of data and computation required for large language models (LLMs), along with the significant capital investment needed to train and deploy them, necessitates the exploration of optimization techniques throughout the LLM lifecycle. In this issue, we focus on potential improvements during the pre-training phase, as this is the most resource-intensive step, involving a vast amount of data and sensitivity to architectural design. Here are some techniques you can employ to improve LLM performance and efficiency: 1. Quantization: Quantization aims to reduce the number of bits needed to store these weights by binning floating-point values into lower-precision buckets. This reduces memory usage with minimal impact on performance. Small precision losses are acceptable as long as the model’s performance is within the required levels. For instance, a weight value like 3.1457898 could be quantized to 3.1458 using a scheme that retains four decimal places. Such a scheme might lead to slight changes (during the backward pass of the training step, for example, a higher margin of error) while computing loss or while updating weights. Take, for instance, 4-bit quantization, which uses small bins where the density of weights is higher and fewer larger bins for weights away from the mean. The 4-bit float representation employs an intelligent approach based on the distribution of model weights. Most weights tend to cluster near zero, with minor differences requiring higher precision, while fewer weights have larger values. To accommodate this, asymmetric binning is used: smaller bins are allocated for values near the mean to maintain precision, while fewer larger bins handle outliers further from the mean. 2. Mixed precision: This is another technique to reduce memory and computational demands without sacrificing significant accuracy. These methods combine different numerical formats, such as float16, int8, and more, to optimize efficiency and performance during training or inference. 3. Data efficiency: Large datasets are costly to process, and redundant or noisy data can negatively impact model performance. Therefore, data efficiency techniques can be applied to achieve high model accuracy and generalization with a reduced or optimized dataset. This process includes filtering data for quality, reducing redundancy, and applying sampling techniques to emphasize high-value samples. 4. Sparse attention: Instead of computing attention weights for every pair of tokens in the input sequence, sparse attention focuses only on a subset of tokens, exploiting patterns in the data or task-specific properties. To put things into perspective, think about decoder-only architectures like GPT trained with an auto-regressive language objective. Such an objective puts a constraint on the attention layer to be causal, and thus, only the lower triangular attention matrix is useful (but the computation is still done for the whole matrix). Different architectures leverage specific patterns, like local or strided attention mechanisms, to bring in efficiency in computation time. 5. Flash attention: Flash attention takes the route of hardware-based improvements and efficiencies to compute attention scores. There are two popular techniques for sparse attention: Kernel fusion and Tiling. Kernel fusion reduces the number of I/O operations by combining all steps (elementwise operations, matrix multiplication, softmax, etc.) into a single read-write operation. This technique is pretty effective during inference. Tiling, on the other hand, breaks down the overall attention calculation into smaller and manageable groups of operations that fit into fast and low-latency GPU memory. For instance, instead of computing softmax across the entire attention matrix at once, FlashAttention computes it over smaller chunks in a numerically stable and tiled fashion, thus making use of faster memory without the need to store a large matrix. 6. Mixture of Experts (MoE) architecture: MOE is an advanced architecture designed to leverage a subset of components (or experts) rather than the whole architecture itself, thereby achieving higher scalability and efficiency. The Experts in this architecture are independent modules or blocks of the network, where each can be trained to specialize in a specific task. While the Router is a module that learns to select which experts to leverage (or activate) for a given input based on different criteria. The Router itself can be a neural network. 7. Efficient architectures: There are a number of different patterns and techniques that have been developed and leveraged by different architectural improvements over the years. Some of the popular architectures are Linformer, Reformer, and Big Bird. Apart from pre-training optimizations, there are other techniques as well, such as fine-tuning and improvements in inference time. More recently, the availability and popularity of small language models and specialized hardware and frameworks has also contributed to significant improvements in the overall efficiency of resource-constrained environments. Liked the Insights? Want to dig in deeper? If you wish to learn more about these techniques or wish to dive deep into foundational aspects of the LLM ecosystem, you can check out the book, Generative AI with Python and PyTorch, Second Edition, by Joseph Babcock and Raghav Bali. BUY NOW 📈LATEST DEVELOPMENT Let’s kick things off with the top stories of the week. China is aiming for the top spot in the AI race with DeepSeek V3's latest release DeepSeek just released -V3-0324, claiming a major boost in reasoning, front-end development capabilities, and smarter tool use. The release positions DeepSeek as a serious contender to models like Code Llama and Codex. You can try out the open-source weights from this HuggingFace card. Publishers claim AI-Search is an internet takeover, Pichai defends it as an innovation In a podcast with Nilay Patel (Editor-in-Chief of The Verge), Google CEO Sundar Pichai shared candid thoughts on AI’s impact on the internet. He defended AI-generated search results amid backlash, insisting they won’t kill the open web. As Google walks a tightrope between innovation and publisher outrage, Pichai expressed confidence that AI will ultimately “enhance,” not erase, human content. He dodged revenue concerns but acknowledged the risks of unchecked AI growth. Catch the full conversation here. Apple’s branding power move with iOS26 A Bloomberg report says that Apple is set to revamp its OS branding game at WWDC-2025. The rebranding will sync all platforms with the upcoming 2026 launch year, setting the stage for a unified, modernized software identity with iOS 26, macOS 26, and watchOS 26. SamA and Ive team up for AI-first products OpenAI is collaborating with design icon Jony Ive and his firm LoveFrom to craft AI-powered products. Jony Ive, Scott Cannon, Evans Hankey, and Tang Tan led io team will collaborate closely with Open AI’s research and engineering teams, with LoveFrom leading design and creative responsibilities. Their goal: to recapture the magic, creativity, and wonder of early Apple-era technology. Hear more about their vision in this video. Anthropic inching towards interpretable AI? Anthropic just cracked open the black box of AI thinking with its latest research, Tracing Thoughts. Using a novel method called dictionary learning, researchers mapped how language models like Claude internally form and organize thoughts. They uncovered thousands of hidden features that resemble abstract concepts and reasoning steps. This breakthrough gives us a glimpse into not just what AI predicts—but how it thinks. Dive into this investigative research here. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0}#converted-body .list_block ol,#converted-body .list_block ul,.body [class~=x_list_block] ol,.body [class~=x_list_block] ul,u+.body .list_block ol,u+.body .list_block ul{padding-left:20px} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

We’re running a short audience survey to help decide Rethinking What an AI Newsletter Should Be Over the past few months, AI_Distilled has grown into a community of readers coming from very different parts of the AI ecosystem. As the space continues evolving, we’ve been thinking carefully about what this publication should become going forward and how we can make it more genuinely useful for the people reading it. So we’ve put together a short survey to understand what readers want more of, what feels missing from current AI media, and where we should take AI Distilled next. It should take around 4 minutes to complete, and every response will directly help shape the next phase of the publication. Take Survey Appreciate you taking the time. LLM Expert Insights, Packt *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;display:none;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}.social_block .social-table{display:inline-block!important}}

As AI advances, the focus shifts to how we manage it. AI_Distilled #133: What’s New in AI This Week Otis: The World's First Cinematic AI Experience Forget generic chatbots. Otis is a wise elder on a cinematic porch at sunset, back turned, voice warm, ready to talk through whatever you're carrying. The world's first cinematic AI experience. April 21st on Kickstarter! Learn more This week, AI felt a little closer to the real world. Anthropic’s Mythos model has already pushed banks and governments into defensive mode, while newer, more controlled releases show how carefully these capabilities must now be handled. At the same time, AI is quietly becoming more useful across specific domains, from scientific research to the infrastructure that runs these systems. It’s a reminder that progress isn’t just about smarter models, but also about how safely and effectively we can use them. LLM Expert Insights, Packt LATEST DEVELOPMENT 🛑 Anthropic’s Mythos model raises global alarm over financial system vulnerabilities - A new AI model from Anthropic, dubbed Claude Mythos, has triggered concern among finance ministers and central bankers after demonstrating the ability to identify vulnerabilities across major operating systems, browsers, and financial infrastructure. The model has already prompted discussions at IMF meetings, with governments and banks being given early access to test and secure their systems before public release. Officials warn that while the technology could strengthen cybersecurity, it also lowers the barrier for malicious actors to exploit critical weaknesses at scale. 🛡️ Anthropic releases Claude Opus 4.7 with reduced cyber capabilities amid safety concerns - Anthropic has launched Claude Opus 4.7, a new model positioned as its most capable general-purpose release, but deliberately less powerful in cybersecurity tasks than its controversial Mythos model. The company says it has added safeguards to detect and block high-risk use cases, reflecting growing concerns about how advanced models could expose system vulnerabilities. The move signals a shift toward controlled deployment, as Anthropic tests how to safely scale models with capabilities that may otherwise pose systemic risks. 🧬OpenAI unveils GPT-Rosalind, a model built for life sciences research - OpenAI has introduced GPT-Rosalind, a domain-specific model designed to support scientific workflows across biology, drug discovery, and genomics. The model focuses on tasks such as hypothesis generation, literature synthesis, and experimental planning, aiming to accelerate early-stage research where timelines can stretch over a decade. Currently available as a research preview, GPT-Rosalind reflects a broader push toward specialized AI systems tailored to complex, real-world disciplines like life sciences. 🧪 OpenProtein aims to make AI-driven protein design accessible to biologists - OpenProtein.AI is building a no-code platform that gives researchers access to advanced protein-design models without requiring machine learning expertise. Founded by MIT researchers, the platform allows scientists to generate, test, and optimize protein sequences using AI, helping accelerate drug discovery and biological research. By lowering the barrier to entry, the company is aiming to bring cutting-edge AI tools directly into the hands of biologists and smaller labs. ☁️ Cloudflare launches unified AI inference layer to support multi-model agents - Cloudflare is positioning itself as a unified inference layer for AI agents, allowing developers to access 70+ models across multiple providers through a single API. The platform is designed to handle real-world agent workflows, where tasks are split across different models, while also managing latency, cost, and reliability. With features like automatic failover and centralized usage tracking, the move reflects a broader shift toward infrastructure that can orchestrate complex, multi-model AI systems at scale. We’re thinking about launching something new If you have two minutes, take our quick survey and tell us what you’d actually want to read. It’ll help us build something that’s genuinely worth your time. Take the 2-minute survey 📈EXPERT INSIGHTS Mastering NLP From Foundations to Agents This week’s Expert Insight comes from Mastering NLP From Foundations to Agents by Lior Gazit and Meysam Ghaffari, a guide that moves from core NLP principles to the realities of building and fine-tuning modern AI systems. As teams push to adapt large models for real-world use, the constraint is often no longer ideas, but resources: compute, memory, and cost. In this excerpt, Gazit and Ghaffari walk through Quantized LoRA (QLoRA), a technique that makes it possible to fine-tune large language models efficiently on limited hardware, without sacrificing performance. Understanding QLoRA QLoRA extends the idea of LoRA to enable fine-tuning of LLMs on a single GPU. The core idea is to keep the base model frozen and stored in 4-bit quantized precision, while training the LoRA adapters in higher precision (such as bfloat16 or float16). This achieves two goals simultaneously: >> Drastically reducing memory requirements >> Allowing the adapters to both compensate for quantization error and adapt the model to downstream tasks Let’s analyze how QLoRA compares to standard LoRA in practice, focusing on the trade-offs between memory reduction and model fidelity. We will specifically demonstrate how techniques such as NF4 (4-bit NormalFloat) and paged optimizers allow us to recover the quality of full fine-tuning while significantly lowering the barrier to entry for model adaptation. READ FULL ARTICLE Built something cool? Tell us. Whether it's a scrappy prototype or a production-grade agent, we want to hear how you're putting generative AI to work. Drop us your story at nimishad@packtpub.com or reply to this email, and you could get featured in an upcoming issue of AI_Distilled. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Better systems, bigger stakes AI_Distilled #134: What’s New in AI This Week The latest models are getting better at handling complex work with less input, and they are starting to behave in ways that feel closer to real collaborators than tools. At the same time, the stakes are becoming clearer. Developments like Anthropic’s Mythos are drawing attention from governments and financial institutions, while new models from across the industry are pushing on cost, speed, and capability. Companies are already adjusting how they work in response. It feels less like a steady upgrade to how AI fits into the world. LLM Expert Insights, Packt LATEST DEVELOPMENT 🤖 OpenAI launches GPT-5.5, pushing toward more capable agentic AI systems - OpenAI has introduced GPT-5.5, its most capable model to date, designed to handle complex, multi-step tasks with minimal guidance. The model shows strong gains in areas like coding, research, and tool use, with improved ability to plan, execute, and iterate across workflows while maintaining speed and efficiency. With enhanced safeguards and early enterprise deployment, the release signals a continued shift toward AI systems that act more like autonomous collaborators than passive tools. 🌍 Anthropic’s Mythos model turns AI into a geopolitical flashpoint- Anthropic’s Mythos model has triggered a global scramble among governments and central banks after demonstrating the ability to uncover critical vulnerabilities across financial systems and infrastructure. Access to the model is tightly controlled, with most countries excluded, turning it into a strategic asset and raising concerns about unequal visibility into emerging cyber risks. The episode highlights a deeper shift: as AI capabilities advance, they are starting to function less like product launches and more like geopolitical leverage points with real security implications. ⚙️ DeepSeek previews V4 model, reinforcing China’s push for low-cost AI leadership- Chinese AI startup DeepSeek has released a preview of its V4 model, building on the disruption caused by its earlier low-cost, high-performance systems. The new model emphasizes strong agent capabilities and lower inference costs, while remaining open-source and optimized for local deployment. With support for domestic chips and growing competition within China, V4 signals a broader shift toward AI sovereignty and cost-efficient alternatives to Western models 🎧 xAI launches Grok Voice Think Fast 1.0 for real-time enterprise voice agents- xAI has introduced Grok Voice Think Fast 1.0, a new voice model designed for real-time, multi-step workflows across customer support, sales, and enterprise applications. The model focuses on low-latency responses, accurate data capture, and reliable tool use in noisy, real-world environments, with early deployments already handling complex support and sales interactions at scale. The release highlights a growing shift toward AI agents that can operate autonomously in live, high-stakes conversations. 📉 Tech layoffs deepen as Meta and Microsoft double down on AI investments- Meta and Microsoft are cutting thousands of jobs while ramping up spending on AI, with Meta planning to reduce its workforce by around 10% and Microsoft offering voluntary exits to a significant portion of employees. Executives point to rising productivity from AI as a key factor, with some claiming that tasks once handled by large teams can now be completed by far fewer people. The moves highlight a growing shift: as companies invest heavily in AI infrastructure and capabilities, workforce structures are beginning to change alongside it. We’re thinking about launching something new If you have a minute, take our quick survey and tell us what you’d actually want to read. It’ll help us build something that’s genuinely worth your time. Take the survey 📈EXPERT INSIGHTS RAG-Driven Generative AI, Second Edition This week’s Expert Insight comes from the second edition of RAG-Driven Generative AI by Denis Rothman, a practitioner who has spent decades building AI systems in real-world enterprise settings. This edition focuses on how RAG is evolving from simple experiments into production-ready systems that work with enterprise data at scale. In this excerpt, Rothman breaks down the RAG ecosystem into its core parts and explains how they fit together. The RAG Ecosystem RAG-driven generative AI is a framework that can be implemented in many configurations. However, the RAG framework runs within a broad ecosystem, as shown in Figure 1.3. No matter how many retrieval and generation frameworks you encounter, it all boils down to the following four domains and the critical questions that accompany them: > Data: Where is the data coming from? Is it reliable? Is it sufficient? Crucially, in the MAS-RAG era, does the data stay within the secure corporate trust boundary? > Storage: How is the data going to be stored? In the traditional approach, data was fragmented between SQL databases and external vector stores. In the modern approach, we ask: Can we store vectors alongside business data in a single converged database? > Retrieval: How will the correct data be retrieved? Will we use simple keyword matching (Naïve) or integrated vector search (Advanced)? > Generation: How will the appropriate generative AI model be selected? How will we securely pipe the retrieved private data into the model? READ FULL ARTICLE View the latest HubSpot Developer Platform updates in Spring Spotlight See what's new for the HubSpot Developer Platform! Ship faster with AI coding tools like Cursor, Claude Code, and Codex. Build MCP-powered AI connectors, run serverless functions with support for UI extensions, and use date-based versioning to streamline roadmap planning. Learn more Built something cool? Tell us. Whether it's a scrappy prototype or a production-grade agent, we want to hear how you're putting generative AI to work. Drop us your story at nimishad@packtpub.com or reply to this email, and you could get featured in an upcoming issue of AI_Distilled. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;display:none;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}.social_block .social-table{display:inline-block!important}}

AI dependence and synthetic influence raise new questions AI_Distilled #136: What’s New in AI This Week Get Tickets Researchers are now questioning whether heavy reliance on AI tools could weaken independent thinking, while AI-generated influencers and automated marketing systems are making it harder to separate expertise from synthetic persuasion. That tension between autonomy and control sits at the center of this week’s Expert Insight. The excerpt explores how modern AI agents are structured internally, particularly the separation between an agent’s persistent identity and the tasks it is tasked with performing. As agents become more embedded into real workflows, those architectural choices are starting to matter far beyond prompt engineering experiments. LLM Expert Insights, Packt LATEST DEVELOPMENT 🧠 Heavy AI dependence may weaken independent thinking, researchers warn -A study by researchers from MIT, Carnegie Mellon, Oxford, and UCLA found that people using AI assistants to solve reading and maths problems completed tasks faster but showed lower engagement with critical thinking and problem-solving processes. The findings raise concerns that growing reliance on AI tools could gradually reduce persistence and independent reasoning skills over time. ⚡ Anthropic doubles Claude usage limits after major SpaceX compute deal -Anthropic has expanded usage limits for Claude Code and its API after signing a compute partnership with SpaceX that gives it access to more than 220,000 NVIDIA GPUs at the Colossus 1 data center. The announcement highlights how competition in AI is increasingly shifting from model capabilities alone to securing massive infrastructure and compute capacity at scale. 🏋️ AI-generated fitness influencers push misleading transformation claims online - Google has developed TurboQuant, a compression method that reduces AI working memory requirements by up to six times without affecting performance. The advance could significantly lower infrastructure costs and enable more powerful models to run efficiently, though it remains at an early stage. 🛠️ Tools worth trying this week - From AI-powered email signature builders to branding assistants that generate polished HTML-ready designs in minutes, these tools show how generative AI is quietly reshaping even the most routine parts of digital work. If you want to experiment with lightweight but practical AI utilities, these are worth a look. We’re thinking about launching something new If you have a minute, take our quick survey and tell us what you’d actually want to read. It’ll help us build something that’s genuinely worth your time. Take the Survey 📈EXPERT INSIGHTS 30 Agents Every AI Engineer Must Build In this week’s Expert Insight, Imran Ahmad, author of 30 Agents Every AI Engineer Must Build, explores one of the foundational ideas behind modern agent engineering: the separation between an agent’s persistent identity and its real-time tasks. The two-layer prompt architecture: System and user prompts One of the most foundational innovations in agent design is the two-layer prompt architecture, which distinctly separates an agent's core identity from its real-time instructions. This layered design, consisting of the system prompt and the user prompt, establishes a clear division of responsibilities, drawing inspiration from classical software principles such as separation of concerns and abstraction layers. A helpful analogy is that of an agent functioning as a diplomat: the system prompt defines the diplomat's country, values, and code of conduct; the user prompt is the current negotiation or message they are handling. The diplomat must respond fluidly, but always in alignment with national policy. In multi-agent scenarios, this diplomat analogy extends across agent boundaries. When one agent passes a task or data payload to another, it is effectively handing off a "diplomatic brief": the receiving agent's system prompt must re-establish persona, authority scope, and operational constraints for the new context. Without explicit role-passing in the handoff protocol, the receiving agent may inherit ambiguous instructions or combine roles across agents. Well-designed multi-agent architectures, therefore, encode the PTCF components not just in each agent's internal system prompt but also in the inter-agent message schema, ensuring that every communication boundary preserves the constitutional clarity that the framework provides. Together, these two layers form what we might call the agent's prompt contract: > System prompt: How the agent behaves > User prompt: What the agent should do Read Full Article Build and test native paywalls in seconds Turn a prompt into a complete native paywall with RevenueCat’s Paywalls AI Editor Update copy, adapt designs for dark mode, and launch A/B tests without waiting for the next sprint.Use free up to $2.5k monthly tracked revenue. 96,000+ apps trust RevenueCat. Learn More Built something cool? Tell us. Whether it's a scrappy prototype or a production-grade agent, we want to hear how you're putting generative AI to work. Drop us your story at nimishad@packtpub.com or reply to this email, and you could get featured in an upcoming issue of AI_Distilled. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;display:none;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}.social_block .social-table{display:inline-block!important}}

From Digit’s delivery test to Gemini 2.5’s native audio and ChatGPT-powered productivity—this week’s AI_Distilled #98: What’s New in AI This Week Join the live "Building AI Agents Over the Weekend" Workshop starting on June 21st and build your own agent in 2 weekend. In this workshop, the Instructors will guide you through building a fully functional autonomous agent and show you exactly how to deploy it in the real world. BOOK NOW AND SAVE 35% Use Code AGENT25 at checkout Spots are limited. Book now to SAVE 35% (Valid for till 8th June 2025) This month is buzzing with AI innovation—from can’t-miss conferences to game-changing GenAI use cases. Whether you're looking to level up your skills, explore new tools, or stay ahead of the curve, we've got you covered. LLM Expert Insights, Packt In today's issue: 🧠 Expert Deep Dive: Valentina Alto explores real-world GenAI use cases—from code and content to campaigns and daily life. 📅 June Conference Watch: Your curated guide to the top AI/LLM conferences this month—CVPR, ICML, ACL, and more. 🎯 Productivity Reimagined: From GTM strategy to custom workouts, see how ChatGPT reshapes personal and professional workflows. 🔊 Gemini 2.5 Gets Audio: Google DeepMind’s latest model understands tone, languages, and screen-shared content. 📦 Amazon’s Humanoid Robot: Digit enters delivery trials—redefining warehouse automation and last-mile logistics. 🔐 OpenAI Boosts Security: A new vulnerability disclosure framework sets industry standards for AI integrity. 🚫 DeepSeek Faces Criticism: China’s newest model sparks global concern with aggressive political censorship. ⚡ Nvidia Dominates MLPerf: Blackwell GPUs set new training records, proving unmatched performance in AI workloads. 📈UPCOMING EVENTS JUNE'S MUST ATTEND AI/LLM CONFERENCES Breakthroughs in AI are made possible through years of study, experimentation, and research that eventually shape the mainstream. Whether you're a researcher pushing the boundaries of machine learning, a developer building with generative AI, or a leader shaping enterprise strategy, this handpicked list of the top conferences in 2025 will help you stay connected to the pulse of innovation. 1. CVPR 2025 – IEEE/CVF Conference on Computer Vision and Pattern Recognition Dates: June 11–15, 2025 Location: Music City Center, Nashville, TN, USA Cost: In-person - General: $900; Student: $810; IEEE/CVF Members ($900 for professionals, $675 for students) Nature: Virtual - General: $215; Student: $125; IEEE/CVF Members ($180 for professionals, $100 for students) Focus: Computer vision, multimodal AI, LLMs in vision tasks Website: CVPR 2025 Conference 2. ICLAD 2025 – IEEE International Conference on LLM-Aided Design Dates: June 26–27, 2025 Location: Paul Brest Hall, Stanford University, Stanford, CA  Cost: In-person only - General: $600; Student: $410; IEEE/CVF Members ($500 for professionals, $350 for students) Focus: Utilizing large language models to enhance design processes in circuits, software, and computing systems Website: International Workshop on LLM-Aided Design 3. ICML 2025 – International Conference on Machine Learning Dates: July 13–19, 2025 Location: Vancouver Convention Center, Vancouver, Canada Cost: In-person - General: $1365; Student: $1030 Nature: Virtual - General: $275; Student: $200 Focus: Machine learning theory and practice, generative AI, LLMs Website: ICML 2025 Conference 4. ACL 2025 – 63rd Annual Meeting of the Association for Computational Linguistics Dates: July 27 – August 1, 2025 Location: Vienna, Austria Cost: In-person - General: $1125; Academic: $800; Student: $425 + ACL Membership fee ($100 for professionals, $50 for students) Nature: Virtual: - General: $550; Academic: $400; Student: $250 + ACL Membership fee ($100 for professionals, $50 for students) Focus: Natural language processing, large language models, language generation Website: ACL 2025 5. NeurIPS 2025 – Conference on Neural Information Processing Systems Dates: December 2–7, 2025 Location: San Diego Convention Center, San Diego, CA, USA Cost: In-person - General: $1000; Academic: $800; Student: $375 Nature: Virtual - General: $275; Academic: $200; Student: $50 Focus: Advanced ML research, LLMs, multimodal AI Website: NeurIPS 2025 Conference EXPERT INSIGHTS FROM TEXT TO TECH: THE MANY USE CASES OF GENERATIVE AI The hype around GenAI and how it enhances productivity shows no signs of slowing down. Just as previous generations shifted from Xeroxing to Googling, we now find ourselves firmly in the era of “Ask ChatGPT.”. GenAI finds its applications in various fields, such as image synthesis and text generation to music composition, marketing content, data analysis, coding, and countless other tasks that, until recently, required specialized expertise. In this issue, we spotlight just a few of the many real-world applications of GenAI, using OpenAI’s ChatGPT as our lens. Here are four use cases from one of our best-selling books, Practical Generative AI with ChatGPT, written by our star author Valentina Alto. 1. Daily assistant: ChatGPT is an excellent tool for boosting your day-to-day activities, such as grocery shopping, meal planning, and workouts, among many other tasks. Take, for example, the following prompt: Generate a 75’ workout routine for strength training. My goal is increasing my overall strength and also improving flexibility. I need a workout for the upper body only divided by the muscle group. Make it in a table format with # of reps and # of series. Make sure to incorporate some rest as well. Here is a sample workout plan that ChatGPT might generate for you: 2. Creating content: You can use ChatGPT to craft emails, create social media posts, write blogs and articles, assist with proofreading, perform translations, analyze documents, or even adjust the tone of your content: whether you want it to be formal, quirky, casual, or sarcastic. Take a look at ChatGPT’s sarcastic translation of an Italian text: 3. Coding assistant: The primary capability you should leverage is ChatGPT’s code generation. From writing a simple function to creating the skeleton of a game, ChatGPT can provide enough building blocks to get started. You can also use it to suggest code optimizations, explain errors, and debug your existing code. Additionally, it can help generate documentation, improve code explainability, and even assist in understanding the structure of a neural network. Take, for example, the following CNN model: If you ask ChatGPT to explain this model, it may respond as follows: 4. Design marketing campaigns: Suppose you have a new product and need a go-to-market (GTM) strategy. You can ask ChatGPT to help you draft an initial plan. Then, by iteratively refining your prompts, you can request suggestions for the product name, marketing hook, target audience research, unique value proposition, sales channels, pricing, SEO keywords, and more. You can even ask it to generate product launch posts. Here are some of the prompts Valentina experimented with in her book while developing a GTM strategy for eco-friendly socks. Generate 5 options for a catchy product line name Generate 3 slogans for the “GreenStride” name. They should be motivating and concise. What kind of target audience should I address with the promotion of GreenStride socks product line. What could be the best channel to reach the segments identified above Give me three concise suggestions on how to make my socks line GreenStride outstanding and unique in a competitive market Generate a product description (max 150 words) for GreenStride socks line using unique differentiator you listed above. It should be attention-grabbing and effective, as well as SEO optimized. List also the SEO keywords you used to finish. What could be the fair price of my socks line I want to generate an Instagram post to announce the launch of GreenStride socks. Write a post (max 150 words) including the unique features and differentiators mentioned above, as well as relevant hashtags. Liked the Insights? Want to dig in deeper? Beyond the four use cases we’ve spotlighted in this issue, the book Practical Generative AI with ChatGPT, by Valentina Alto, introduces generative AI and its applications, focusing on OpenAI’s ChatGPT. It covers prompt engineering, daily productivity use cases, domain-specific applications for developers, marketers, and researchers, and the creation of custom GPTs using the GPT Store, enabling specialized assistants without coding, powered by personalized instructions and tools. BUY NOW 📈LATEST DEVELOPMENT Let’s get right into it. Google DeepMind Introduces Gemini 2.5 with Native Audio Capabilities Google DeepMind has launched Gemini 2.5, now capable of processing real-time audio and video. The model can interpret screen-shared content, respond to tone and background noise, and supports over 24 languages, making it more contextually aware and interactive than ever before. Amazon to Test Humanoid Robots for Package Deliveries The Information has reported that Amazon is preparing pilot tests of Agility Robotics' bipedal humanoid robot, Digit, for use in logistics and package handling. Designed to work safely in spaces designed for humans, Digit is expected to automate repetitive warehouse tasks and even assist in last-mile delivery operations. OpenAI Launches Coordinated Vulnerability Disclosure Framework OpenAI has introduced an “Outbound Coordinated Vulnerability Disclosure” policy to responsibly report security issues it uncovers in external systems. This move aims to bolster security standards and transparency across the tech ecosystem. DeepSeek’s New AI Sparks Free Speech Concerns Chinese AI developer DeepSeek has triggered global criticism for its model’s extreme content filtering. Users attempting to query politically sensitive topics, like Tiananmen Square or Taiwanese independence, are met with complete denials, spotlighting a stark divide in global AI moderation norms. Nvidia Blackwell Chips Dominate New MLPerf Benchmarks Nvidia’s Blackwell GPUs dominated the latest MLPerf training benchmarks, delivering double the performance of previous H100 chips. These results highlight Blackwell’s efficiency in training large AI models with fewer GPUs, reduced energy use, and lower costs, solidifying Nvidia’s leadership in AI hardware and accelerating industry-wide adoption of its new architecture. Kubernetes for Generative AI Solutions 40% Off on eBook + 20% Off on Paperback for the next 48 hours 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0}#converted-body .list_block ol,#converted-body .list_block ul,.body [class~=x_list_block] ol,.body [class~=x_list_block] ul,u+.body .list_block ol,u+.body .list_block ul{padding-left:20px} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Agentic Engineering is live. Agentic Engineering Is Now Live I'm interrupting your very busy schedule for an announcement that you might’ve already seen coming. Our newsletter, Agentic Engineering, has finally kicked off. We sent out surveys. We sent out pre-launch messages. And you’ve all been really supportive while we figured out the logistics of kicking off something new. Across the surveys and interviews, you had one clear frustration. Too much hype around updates, launches, and releases, and very little guidance on what to actually do with all of it. That gap is what Agentic Engineering is for. We will lean on our network of experts who are actively building in this space, so you’re not just getting opinions, but conversations from people making decisions right now, often before they show up on timelines. But enough said! We’ll leave it to you to decide how useful this space is. Subscribe if this sounds like what you’ve been looking for. You can always unsubscribe later. And just to nudge things along (without making it too transactional), early subscribers will receive a free ebook copy of AI Agents in Practice by Valentina Alto! Tanya, Agentic Engineering Join Agentic Engineering and Grab Your Free Copy 📈EXPERT INSIGHTS A preview of Agentic Engineering Something Maxime Labonne said during one of our earlier roundtables stuck with me because it runs counter to how people think about small models. The common assumption is that small models are simply the cheaper version of frontier models. Same idea, lower cost. But Maxime’s experience has been that they’re often harder to work with. Not because they’re worse, but because they expose problems that larger models can often hide. When most people build AI systems today, they’re testing them with frontier models. Those models are smart enough to compensate for weak prompts, incomplete logic, or edge cases that nobody thought about. Small models don’t give you that luxury. As Maxime put it, they can fail on surprisingly basic tasks, and when they do, entire workflows can break. Read the Full Article *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;display:none;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

It's moving deeper into the fabric of technology AI_Distilled #140: What’s New in AI This Week Avail 40% Off Today LLM Expert Insights, Packt LATEST DEVELOPMENT 🎮 Pokémon Go scans helped train AI now being explored for military drones - Location scans voluntarily submitted by Pokémon Go players have been used to train AI models that help machines understand and navigate physical environments. Following a partnership between Niantic Spatial and drone software company Vantor, the technology is now being explored for use in GPS-denied environments, raising fresh questions about how consumer-generated data may ultimately be used in military and defence applications. 🏗️ Mistral bets on agents, infrastructure, and custom AI chips - Mistral CEO Arthur Mensch says enterprise AI adoption is still in its early stages, with significant value yet to be unlocked as organizations adapt to agentic workflows. He also revealed that the French AI startup is exploring the development of its own chips, signaling ambitions to control more of the AI stack as it expands beyond models into infrastructure and enterprise deployment. 👷 Jeff Bezos pushes back on AI job-loss fears - Jeff Bezos argues that AI-driven productivity gains will create new industries, products, and jobs rather than trigger mass unemployment. Speaking about his AI startup Prometheus, Bezos said the bigger long-term challenge may be labor shortages, as AI accelerates innovation across sectors such as manufacturing, aerospace, semiconductors, and energy. 🎓 Anthropic launches $150 million AI fellowship program - Anthropic has unveiled Claude Corps, a $150 million initiative that will train and place 1,000 early-career professionals at nonprofits across the U.S. The program aims to help organizations adopt AI tools while equipping participants with practical AI skills, reflecting growing efforts to distribute the benefits of AI more broadly amid concerns about workforce disruption. ⚡ Google unveils DiffusionGemma for faster AI text generation - Google has released DiffusionGemma, an experimental open-source model that uses diffusion techniques instead of traditional token-by-token generation, enabling text generation speeds up to four times faster on dedicated GPUs. While not intended to replace conventional LLMs for quality-critical applications, the model offers a glimpse into alternative architectures designed for real-time, interactive AI workflows. 📈EXPERT INSIGHTS OpenClaw + LangGraph Playbook It’s easy to build an agent that talks. Building one that remembers things, sends messages, runs on a schedule, and generally makes itself useful is a different challenge. That’s where LangGraph and OpenClaw make an interesting combination.Let’s build one. The first step is creating the primary agent and establishing its responsibilities through a system prompt. import os from datetime import datetime from langclaw import Langclaw from langclaw.gateway.commands import CommandContext # Initialize the master agent application interface app = Langclaw( system_prompt=( “## Corporate Intelligence Agent\n” “You are a corporate intelligence analyst. You track market trends “ “and draft precise outreach sequences based on current events.\n” “Delegate deep multi-source research tasks to the web-researcher subagent.” ), ) Tools allow the agent to access capabilities beyond language generation. In this example, the agent can retrieve market intelligence data. Read The Full Article Built something cool? Tell us. Whether it's a scrappy prototype or a production-grade agent, we want to hear how you're putting generative AI to work. Drop us your story at nimishad@packtpub.com or reply to this email, and you could get featured in an upcoming issue of AI_Distilled. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;display:none;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}.social_block .social-table{display:inline-block!important}}

Yann LeCun’s warning, Anthropic’s expansion, and a deeper look at agent evaluation AI_Distilled #141: What’s New in AI This Week Vector memory for AI agents in air-gapped, regulated, and offline environments VectorAI DB delivers sub-15ms retrieval for agent memory and RAG pipelines on your own infrastructure. On-premises, at the edge, or air-gapped. Native support for LangChain, LlamaIndex, and Hugging Face. Free Community Edition available. Get Started for Free LLM Expert Insights, Packt LATEST DEVELOPMENTS 📉 AI pioneer warns of industry bubble as costs outpace revenues - AI researcher Yann LeCun has criticized Elon Musk’s xAI as a struggling competitor in the race for frontier AI while warning that the broader industry risks a “big bubble explosion” if leading labs fail to reduce costs or raise prices. LeCun argued that today’s AI services remain heavily subsidized by investors and suggested that more advanced AI systems may ultimately require new architectures beyond large language models. 🧠 MIT gives robots a memory that works more like ours - MIT researchers have developed a new memory framework that allows robots to remember objects, locations, and past observations using natural language, enabling them to answer questions such as “Where did I leave my wallet?” By combining 3D mapping with AI-generated descriptions, the system could help future robots navigate complex environments and collaborate more naturally with humans. 🌏 Microsoft becomes the primary gateway for OpenAI models in China - While OpenAI and Anthropic have largely stayed out of the Chinese market, Microsoft has emerged as the main supplier of OpenAI’s models to major Chinese technology companies through Azure. The arrangement highlights Microsoft’s unique position in the global AI ecosystem, even as concerns grow around model distillation, geopolitical tensions, and the flow of advanced AI capabilities across national boundaries. 🇰🇷 Anthropic expands into South Korea with new office and AI partnerships - Anthropic has opened a Seoul office and announced partnerships with major Korean organizations, including NAVER, Samsung SDS, LG CNS, and Nexon, as demand for Claude continues to grow across the region. The company also signed an agreement with South Korea’s Ministry of Science and ICT to collaborate on AI safety, cybersecurity, and responsible AI adoption. ⚖️ Study highlights why AI still struggles to moderate online hate speech - New research shows that leading AI moderation systems often disagree on what constitutes hate speech, producing inconsistent results across demographic groups and content types. While AI can detect explicit abuse at scale, researchers say it still struggles with context, sarcasm, coded language, and reclaimed terms, underscoring the challenges of relying on automated systems for online content moderation. Claude is currently the most powerful tool of 2026. Yet almost no one knows how to actually use them. Our expert mentors have condensed 800+ hours of Claude research, articles, YouTube content and real-world practice into a focused 16-hour curriculum. Join the 2-Day Claude AI Mastery Workshop: a live, end-to-end deep dive into Claude plus 10+ AI tools, LLMs and workflows. You will learn how to: - master Claude's three modes : Chat, Cowork and Code. - Set up Skills, Connectors and Plug-ins to automate your desktop, Notion and files. - Vibe code apps and dashboards without writing code & 10+ AI tools and workflows that pair with Claude. 🧠 Saturday & Sunday 🕜 10 AM – 7 PM EST Register NOW! 📈EXPERT INSIGHTS Why a Good Answer Doesn’t Mean a Good Agent During a time when AI conversations are often louder than they are useful, Ammar Mohanna, PhD, brings a refreshing perspective. His career has moved fluidly between academia and industry, from teaching advanced AI courses at the American University of Beirut to advising teams on turning machine learning ideas into systems that can be trusted. He is also known for his candid take on the current AI landscape, especially the gap between meaningful engineering and what he often calls AI slop. In this conversation, Ammar challenges one of the most common assumptions in agent development: that a correct answer is evidence of a successful agent. He explains why reliability lies in the path an agent takes, not just in the result it produces, and why evaluation must evolve from output scoring to a discipline that measures behaviour and trustworthiness in production. Most teams think they’re evaluating agents, but they’re actually not. Where do you see the biggest illusion of evaluation today? The biggest illusion is that teams think they are evaluating an agent when they are only evaluating the final answer. That works well for a chatbot. But an agent is different. It plans, chooses tools, passes arguments, reads observations, retries, stops, and sometimes takes action. A final-answer score hides most of the actual failure surface. An agent can produce a good-looking answer after calling the wrong tool, wasting ten steps, misreading a tool result, or ignoring a failed call. From the outside, the answer may look acceptable. From a reliability perspective, the run is not acceptable. So the illusion is: “the answer looked right, therefore the agent worked.” However, what you need to know is whether the path was valid, efficient, grounded, and safe. Read the Full Interview on Substack Most Claude Code content focuses on prompts and quick wins. This workshop explores what comes next. Join Sam Keen, former engineer at AWS, Lululemon, and Nike, to learn how high-performing teams use structured context, reusable skills, workflow memory, and guardrails to get more consistent results from Claude Code. 🎟️ Exclusive for AI Distilled subscribers: Get 60% off with code AI60. Limited to the first 10 sign-ups. Register Now Built something cool? Tell us. Whether it's a scrappy prototype or a production-grade agent, we want to hear how you're putting generative AI to work. Drop us your story at nimishad@packtpub.com or reply to this email, and you could get featured in an upcoming issue of AI_Distilled. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;display:none;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}.social_block .social-table{display:inline-block!important}}

This week’s Expert Insight explores the agent spectrum AI_Distilled #138: What’s New in AI This Week SUBSCRIBE AI is entering an interesting phase. The biggest questions are no longer about what models can generate, but what systems can be trusted to do. That theme runs through this week’s stories, from AI-designed vaccines and industrial AI partnerships to debates around governance and autonomy. It also sits at the center of this week’s Expert Insight, which explores the spectrum of today’s agents and why not all “agents” are actually the same thing. LLM Expert Insights, Packt LATEST DEVELOPMENT 🧬 AI-designed vaccine enters human trials in world-first study- Researchers at the University of Cambridge have developed what they describe as the first AI-designed vaccine to enter human trials, using AI to create a “super-antigen” capable of protecting against entire families of viruses. The approach could pave the way for universal vaccines against coronaviruses, influenza, and future pandemic threats. 🛑 Anthropic co-founder calls for an AI “brake pedal” - Anthropic co-founder Jack Clark has warned that AI systems are approaching a point where they could increasingly develop without human input, arguing that governments need new regulatory frameworks to maintain control. His comments come as AI capabilities accelerate and concerns grow around economic disruption, autonomous systems, and long-term governance. 📈 Investors look to Asia for the next wave of AI growth - Investment strategists are increasingly pointing to Taiwan and South Korea as the next major beneficiaries of the AI boom, citing their central role in semiconductor and AI infrastructure supply chains. With valuations still below many U.S. AI stocks, some investors see emerging markets as offering significant upside in the next phase of AI-driven growth. 🏭 Hitachi and Intel partner to advance industrial AI and digital infrastructure - Hitachi and Intel have announced a strategic collaboration to accelerate AI adoption across manufacturing, energy, mobility, and other critical industries. The partnership will focus on areas including physical AI, edge computing, quantum technologies, and factory automation, to build more intelligent and resilient industrial infrastructure. 📈EXPERT INSIGHTS A preview of Agentic Engineering The current AI ecosystem has developed a habit of describing wildly different systems with the exact same word: agents. A retrieval pipeline that reformulates search queries, a workflow assistant that schedules meetings, and a system capable of coordinating multi-step operational decisions with minimal supervision now all routinely get discussed under the same umbrella. And I think that’s why the conversation around agents sometimes feels simultaneously overcomplicated and vague. In reality, though, these systems are operating at very different levels of autonomy. So, I wanted to take a new angle: the different kinds of agent organizations are building, and how capabilities change as systems move from retrieval into action and eventually toward autonomy. Read The Full Article Built something cool? Tell us. Whether it's a scrappy prototype or a production-grade agent, we want to hear how you're putting generative AI to work. Drop us your story at nimishad@packtpub.com or reply to this email, and you could get featured in an upcoming issue of AI_Distilled. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;display:none;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}.social_block .social-table{display:inline-block!important}}

What this surprising alliance means for GPU scale, speed, and the future of foundational models. AI_Distilled #99: What’s New in AI This Week Your Exclusive Invite for the World’s first 2 day AI Challenge (usually $895, but $0 today) 51% of companies have started using AI Tech giants have cut over 53,000 jobs in 2025 itself And 40% of professionals fear that AI will take away their job. Join the online 2-Day LIVE AI Mastermind by Outskill - a hands-on bootcamp designed to make you an AI-powered professional in just 16 hours. Usually $895, but for the next 48 hours you can get in for completely FREE! 📅Kick off Call & Session 1- Friday (10am EST- 1pm EST) 🧠Sessions 2-5: 🕜Saturday 11 AM to 7 PM EST ; Sunday 11AM EST to 7PM EST All by global experts from companies like Amazon, Microsoft, SamurAI and more. And it’s ALL. FOR. FREE. 🤯 🚀 🎁 You will also unlock $3,000+ in AI bonuses: 💬 Slack community access, 🧰 Your Personalised AI tool kit, and ⚙️ Extensive Prompt Library with 3000+ ready-to-use prompts — all free when you attend! JOIN NOW - LIMITED FREE SEATS Warm greetings from the AI Distilled team! Here's your freshly baked issue of AI Distilled. With groundbreaking tools and surprise collaborations, this edition is served piping hot. Plus, don’t miss our curated roundup of local AI meetups to keep your network as sharp as your skills. LLM Expert Insights, Packt In today's issue: 🧠 Expert Deep Dive: Shanthababu Pandian shares a blueprint for building scalable, ethical, and adaptive agentic AI systems. 📅 Must-Attend Meetups: From GPU hack weekends to GenAI showcases, here are 5 can’t-miss midsummer AI events across the globe. ⚙️ OpenAI Drops o3-Pro: A high-reasoning model for complex coding, analysis, and real-time search—priced for pros. 🎞️ Meta Goes Multimodal: New AI video editor + V-JEPA 2 pushes Meta’s edge in creative and physical reasoning AI. 🧠 Mistral Debuts Magistral: Their first reasoning-focused model launches alongside Mistral Compute, an enterprise-grade AI infra stack. 🌩️ OpenAI Teams with Google Cloud: Surprise GPU partnership expands OpenAI’s compute scale beyond Azure. 🌍 Google.org Backs Ethical GenAI: $30M accelerator funds nonprofits solving global crises with generative AI. 🔐 EchoLeak Targets Copilot: A zero-click exploit exposes AI’s growing attack surface—Microsoft acts fast. 📈UPCOMING EVENTS MUST ATTEND AI/LLM MEET-UPS Here’s your go-to calendar for this month’s midsummer AI meetups—perfect for networking, learning, and getting hands-on with the latest in generative models, agent frameworks, LLM tooling, and GPU hacking. 1. The Agent – Part 2 Date: June 23, 2025 Location: Cambridge, MA – Boston Generative AI Cost: US $22 Focus: Agent-centric GenAI patterns Website: Meetup Boston 2. Practical AI Monthly Date: June 24, 2025 Location: London – Mindstone AI Cost: Free Focus: Hands-on GenAI use-cases Website: Mindstone London 3. GPU Programming Hack Weekend Dates: June 27–29, 2025 Location: Los Altos, CA – Modular Meetup Cost: Free Focus: Mojo/MAX GPU kernels & PyTorch ops Website: Meetup Los Altos 4. July Mixer & Showcase Date: July 2, 2025 Location: Austin, TX – LangChain AIMUG Cost: Free Focus: LangChain, LLM tooling Website: AIMUG 5. Pizza, Demos & Networking Date: July 9, 2025 Location: Berlin – AI Builders Cost: €5 – €10 Focus: Building with LLMs & GenAI Website: Meetup Berlin What’s stopping you? Choose your city, RSVP early, and step into a room where AI conversations spark, and the future unfolds one meetup at a time. LAST CHANCE - BUY NOW AT 25% OFF EXPERT INSIGHTS - BY SHANTHABABU PANDIAN QUICK UNDERSTANDING OF EFFECTIVE AGENTIC SYSTEM DESIGN Agentic systems, software architectures where autonomous agents act, learn, and interact to achieve goals, are transforming industries from robotics to customer service. These systems, powered by artificial intelligence (AI), enable dynamic decision-making in complex environments. This article provides a concise overview of designing effective agentic systems, focusing on core principles, components, and practical considerations. Shanthababu Pandian, Director- Data and AI, Rolan Software Service What is an Agentic System? An agentic system consists of one or more agents that operate autonomously or semi-autonomously to accomplish tasks. Agents perceive their environment, process information, make decisions, and act, often adapting through the process of learning. Unlike traditional software with fixed rules, agentic systems thrive in dynamic, uncertain settings. Key Characteristics: Autonomy: Agents make decisions without constant human intervention. Reactivity: Agents respond to environmental changes in real-time. Proactivity: Agents pursue goals proactively, anticipating needs. Adaptability: Agents learn from experience to improve performance. Social Ability: Agents collaborate with other agents or humans. Examples include autonomous drones, AI-driven chatbots, or multi-agent systems in logistics optimization. Core Principles of Effective Design Designing agentic systems requires striking a balance between autonomy, efficiency, and reliability. Below are the foundational principles: Core Principles of Effective Design Designing agentic systems requires striking a balance between autonomy, efficiency, and reliability. Below are the foundational principles: Goal-Oriented Design: Define clear, measurable objectives for agents (e.g., “deliver packages in under 30 minutes”). Align agent goals with system-wide outcomes to avoid conflicts in multi-agent setups. Modularity: Build agents with modular components (perception, decision-making, action) for flexibility and easier updates. Example: A robotic agent’s vision module can be upgraded without altering its navigation logic. Robust Perception: Equip agents with sensors or data inputs to accurately interpret their environment. Use redundancy (e.g., multiple sensors) to handle noise or failures. Scalable Decision-Making: Implement decision-making algorithms (e.g., reinforcement learning, rule-based systems) that scale with complexity. Balance computational cost with decision quality—simple heuristics may suffice for some tasks. Learning and Adaptation: Incorporate learning mechanisms (e.g., machine learning models) to adapt to new scenarios. Use online learning for real-time updates and offline training for stability. Coordination in Multi-Agent Systems: Design communication protocols for agents to share information and negotiate. Use centralised (e.g., a coordinator agent) or decentralised (e.g., consensus algorithms) approaches based on system needs. Safety and Ethics: Embed fail-safes to prevent harmful actions (e.g., collision avoidance in drones). Key Components of Agentic Systems An effective agentic system typically includes: Perception Module: Collects data from the environment (e.g., cameras, APIs, user inputs). Processes raw data into actionable insights using techniques like computer vision and natural language processing. Decision-Making Module: Choose actions based on goals and perceived state. Common approaches include rule-based logic, planning algorithms, or AI models like deep reinforcement learning. Action Module: Executes decisions (e.g., moving a robot arm, sending a message). Interfaces with hardware and software actuators. Learning Module: Update agent behaviour based on feedback (e.g., rewards in reinforcement learning). Store knowledge in models or databases for future use. Communication Module (for multi-agent systems): Enables agents to share states, plans, or resources. Utilises protocols such as MQTT or gRPC for efficient data exchange. Practical Considerations Environmental Analysis: Understand the environment’s dynamics (e.g., predictable vs. chaotic) to choose appropriate algorithms. Example: A warehouse robot needs robust navigation in a structured environment, while a chatbot must handle unpredictable user inputs. Resource Constraints: Optimise for computational, energy, or bandwidth limits, especially on edge devices like IoT sensors. Example: Use lightweight ML models for real-time processing on drones. Testing and Validation: Simulate environments to test agent behaviour under diverse scenarios. Use formal verification for critical systems (e.g., autonomous vehicles) to ensure safety. Scalability: Design systems to handle increasing numbers of agents or tasks. Example: A logistics system should support adding more delivery drones without degrading performance. Human-Agent Interaction: Create intuitive interfaces for human oversight and collaboration. Example: A customer service agent should seamlessly escalate complex queries to human operators. Challenges and Solutions Challenge: Unpredictable environments can lead to poor agent performance. Solution: Use robust learning algorithms (e.g., meta-learning) and fallback mechanisms. Challenges: Multi-agent coordination can cause conflicts or inefficiencies. Solution: Implement game-theoretic approaches or swarm intelligence techniques. Challenges: Ethical concerns, like bias in decision-making. Solution: Audit training data and incorporate fairness constraints in models. Real-World Applications Logistics: Multi-agent systems optimise delivery routes (e.g., Amazon’s warehouse robots). Healthcare: AI agents assist in diagnostics or patient monitoring. Gaming: NPCs (non-player characters) act as autonomous agents for immersive experiences. Smart Cities: Agents manage traffic flow or energy distribution. Conclusion Effective agentic system design hinges on clear goals, modular architecture, and robust adaptation mechanisms. By prioritising scalability, safety, and coordination, developers can create systems that thrive in dynamic environments. As AI advances, agentic systems will play an increasingly central role in automating complex tasks, driving efficiency, and enhancing human capabilities. For further exploration, consider open-source frameworks like ROS (Robot Operating System) for robotics or RLlib for reinforcement learning-based agents. Liked the Insights? Want to dig in deeper? Master the art of building AI agents with large language models using the coordinator, worker, and delegator approach for orchestrating complex AI systems Understand the foundations and advanced techniques of building intelligent, autonomous AI agents Learn advanced techniques for reflection, introspection, tool use, planning, and collaboration in agentic systems Explore crucial aspects of trust, safety, and ethics in AI agent development and applications BUY NOW 📈LATEST DEVELOPMENT Here is the news of the week. OpenAI Debuts o3-Pro Model OpenAI has quietly introduced o3-pro, an advanced "high-reasoning" version of its o-series models designed for research, complex analysis, and coding. Featuring real-time web search, Python execution, and multimodal reasoning, o3-pro starts at $20–$80 per million input/output tokens—a tenfold increase over the standard o3. Preliminary tests indicate improved accuracy in science, business, and writing tasks, despite slightly slower response times. Meta Unveils AI Video Editor and Physical Reasoning AI World Model Meta’s new generative AI video editor transforms any ten-second clip into a customizable playground. Now available on the Meta AI app, Meta.ai, and the Edits mobile app, users can upload clips and apply over 50 preset prompts to alter clothing, settings, lighting, or visual styles within seconds. This feature is free for a limited time, and edited clips can be directly shared on Facebook or Instagram. Additionally, Meta unveiled V-JEPA 2, a sophisticated "world model" that enhances robotic and AI agent reasoning capabilities. V-JEPA 2 is trained to recognize patterns in physical interactions, such as the dynamics between people, objects, and their environment. To support community engagement, Meta has open-sourced three new test suites, inviting researchers to rigorously evaluate and accelerate the development of machine common sense. Mistral returns with Magistral Reasoner and Mistral Compute Paris-based Mistral AI has launched Magistral, its first dedicated reasoning model, available in both open-source and enterprise tiers. Magistral prioritizes transparent, step-by-step logical reasoning, deep domain expertise, and extensive multilingual support, directly addressing common criticisms of earlier chain-of-thought models. Complementing this launch, Mistral introduced Mistral Compute, an infrastructure solution providing bundled GPUs, orchestration, and managed services. The offering allows governments, enterprises, and research institutions to operate cutting-edge AI on-premises or within national cloud infrastructures, reducing dependency on U.S.-based cloud providers. OpenAI–Google Cloud Alliance In an unexpected strategic collaboration, OpenAI has partnered with Google Cloud for additional GPU capacity, complementing its existing partnerships with Microsoft Azure and CoreWeave. Finalized in May, this deal helps OpenAI scale rapidly and diversify its supply chain. Google.org Funds Social-Impact Gen-AI for its 2025 GenAI Accelerator program Google.org has selected 20 nonprofits and civic groups for its 2025 Generative AI Accelerator program. Awardees will receive six months of technical mentorship, pro-bono AI expertise, cloud credits, and a portion of a $30 million fund to address critical global issues, from crisis response and children's mental health to combating antimicrobial resistance. Zero-Click EchoLeak Hits Copilot Security researchers at Aim revealed EchoLeak, a novel zero-click exploit targeting Microsoft 365 Copilot. The vulnerability allowed malicious markdown emails to bypass prompt-sanitization, triggering background HTTP requests capable of exfiltrating sensitive data without user interaction. Microsoft swiftly patched the vulnerability before its public disclosure, highlighting emerging security risks associated with increasingly autonomous AI systems. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! That’s a wrap for this week’s edition of AI_Distilled 🧠⚙️ We would love to know what you thought—your feedback helps us keep leveling up. 👉 Drop your rating here Thanks for reading, The AI_Distilled Team (Curated by humans. Powered by curiosity.) *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0}#converted-body .list_block ol,#converted-body .list_block ul,.body [class~=x_list_block] ol,.body [class~=x_list_block] ul,u+.body .list_block ol,u+.body .list_block ul{padding-left:20px} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}