Limitations of generative AI

There are quite a few limitations to generative AI to be aware of. Having an understanding of these limitations also helps to have realistic expectations of what these tools can do for you. We will focus on three: bias, context size, and perceived reasoning versus non-deterministic reasoning.

Bias

The first limitation to be aware of is that when LLMs are being created, they are trained in such a way that they focus on the data they have seen the most, in the order of the words they have seen the most. That means that the response they will generate is based on the common denominator of the source data. You can imagine that these models have seen the phrase “The capital of France is Paris” more often than many other capital cities and countries. Hence, they will most likely complete the prompt, “What is the capital of France?” with the correct answer, “Paris.” This means that the quality of the results of...