@vincent @alex @gme @vincent @Moon @ai what does "understand" mean though? I mean if you really wanted to hammer out an airtight definition. a detailed explanation of *exactly* what did and didn't qualify

is human abstraction a sophisticated form of lossy compression or something more?

is gpt different from a markov chain on a fundamental level or is it just doing the same thing but a lot better?

I do wonder how gpt can make things rhyme since it (presumably) has no idea what things sound like when spoken aloud. it's interesting to think that the relative associations between the tokens in the sequence combined with the broader surrounding context was apparently sufficient to get it outputting stuff that fits such a convoluted pattern (convoluted because the underlying rule for the pattern is based on the spoken word)