Discussion about this post

User's avatar
Russell Hogg's avatar

I always enjoy the double standard where women can be better than men in any number of fields but if men are better than women at something (say at chess) it’s because they are actually worse! Narcissistic self gratification . . .

Expand full comment
SkinShallow's avatar

I keep feeling I'm ignorantly missing something that must be obvious when confronted with opinions of people - including many AI researchers - who suggest LLMs are in any way conscious or even intelligent in remotely human way.

I'm not sure if it's the fact that those people exist largely in the semi autonomous world of text fragments and code themselves and thus believe that this is what world is made of -- but the chess example is an excellent demonstration of a wider fact: LLMs know, in the sense of understanding and being able to reliably manipulate and generate from, ABSOLUTELY NOTHING about the world outside their training data and (if prompted) a quick search of generally available Internet sources. And even their "factual" knowledge -- not understanding which they have none of -- is unreliable, especially as you go down to anything even remotely detailed or niche or lower level.

And I'm saying it as someone who uses LLM to bounce ideas of, and tidy up/structure my own thinking all the time, and effectively. Especially for therapeutic/counselling adjacent purposes for which they have plenty of material to be useful at.

As to writing, it's not just substack. I've read a (good in ideas) piece in legacy media recently that abounded with "it's not X, it's Y" phrasing and dashes so beloved of ChatGPT. And yes, while the thought was there, and valuable, the execution was mediocre LLM slop.

Expand full comment
18 more comments...

No posts