[sethev]

It's kind of an aside in the post, but connecting LLMs and Searle's Chinese Room argument is a brilliant observation. Although there are people who believe LLMs are really thinking, it's mostly confirming that the Turing test wasn't the right way to test this.

[forgetfreeman]

Is it? The observation seems patently obvious if one has even the most superficial understanding of how LLMs work? Why is this not common knowledge?

[elzbardico]

Understanding LLMs require a few pieces of knowledge that are not very common in the industry, not only ML stuff, but I think that having worked in the past with NLP helps a lot.

[forgetfreeman]

I fucked around with dissociate press 30 years ago just to see what would happen if you fed it a combination of random chapters from Alice in Wonderland and the book of Revelations. This hardly represents insider knowledge.