Bowman acknowledges that the reasoning capabilities of LLMs in general are “spotty” and more limited than in people — but he says that they are there, and seem to improve with model size, which indicates to him that future LLMs will be even better.

– Celeste Biever

How smart are large language models? Like defining intelligence, the answer is elusive, but growing in importance as LLMs progress.

Some researchers believe LLMs exhibit reasoning while others say they fall short of understanding anything they generate.

Nature talks with researchers on both sides of the debate.


ChatGPT broke the Turing test — the race is on for new ways to assess AI | NATURE | July 25, 2023 | by Celeste Biever