The gap in shared understanding of LLM capability is widening

A tweet by Andrew Karpathy, the former director of AI at Tesla, highlights a growing disparity between the perceived and actual capabilities of Large Language Models (LLMs). The tweet contrasted the idea that LLMs are 'generative' with the reality that they are 'descriptive'. He also mentioned that LLMs are often mistakenly thought to be able to 'understand' text in the way humans do, but in reality, they are just pattern-matching. This gap in understanding is attributed to the public's lack of exposure to the inner workings of LLMs, as well as the hype surrounding AI technology. The tweet sparked a discussion on Hacker News, with some users sharing their own experiences with LLMs and the limitations they have encountered.