What's the use case of for an "AI" inherently incapable of telling apart high-perplexity text (schizo word salad) from high-perplexity text (conceptually groundbreaking work of genius)?
It's a text generator. It's basically a perfect BS machine in an age where middle managers, labor aristocrats in big companies & women in ordinary jobs have BS jobs doing nothing all day.
>>108954255>bringing up the actual technical limitation of LLMsthe spambots won't touch this one
>>108954255same as with humans
>>108954255Ai won, Luddites lost
>>108955149AI won so bigly we're doing real science here, thoughts VibeGOD? >>108932832
>>108954255Exposing the entire field of sociology as a fraud.
frontier AI models are solving previously unsolved erdos math problems and luddites are still living in 2023 where AI couldn't even code or do mathInsane how many luddite retards browse a fucking technology board lmao
Formal verification, neuromorphics
>>108955085>same as with humansLow-IQ take. The closest human analogue for high perplexity would be struggling to understand what a text is about and where the author is going with it, but as you progress through it, your evaluation can change drastically such that whatever came earlier suddenly makes perfect sense in light of what came later. This doesn't happen with causal transformers that still make up the core of fancy LLMs. Humans also don't have that symmetry between generating and evaluating a text: if you're the author, an opening section that baffles the reader and leaves him wondering where you're going with this makes sense to you because you know exactly where it's going.
>>108956883>Low-IQ take. The closest human analogue for high perplexity would be struggling to understand what a text is about and where the author is going with it, but as you progress through it, your evaluation can change drastically such that whatever came earlier suddenly makes perfect sense in light of what came later.this literally happened to me with both critique of pure reason and being and time.
