The next time you ask ChatGPT or another Generative AI service a question, take note of the order of the words in the answer you get.
The last few words in your query will almost invariably be the first few words of the generated response.
That’s an important process of how large language models (LLMs) works — backwards, evaluating each word in terms of each previous one — pointed out Bruce Momjian, EDB VP and Postgres evangelist, in a talk at FOSDEM 2025, in Brussels, last month.
Using database technology as a launching point, Momjian…