I think this is in fact what’s going on. GPT2 can say things that don’t make sense with a fluent prose style; therefore it is possible to have style and make no sense.
-
Show this thread
-
In fact you can have note-perfect, sophisticated, convincing style without coming one inch closer to making sense.
3 replies 1 retweet 12 likesShow this thread -
Style and sense may just be two disjoint capacities. If I’m speculating I tend to think the “cortical algorithm” is for style, and sense is extrapolated from something striatal or related to motor/kinaesthetic learning.
2 replies 0 retweets 9 likesShow this thread -
Personally it’s frustrating because I have awesome style and just enough sense to keep up, and most of the people I know are sorta “sense-supremacists” who think that skill profile is a hallmark of bad people.
2 replies 0 retweets 8 likesShow this thread -
A lot of people just don’t believe in their hearts that different skills are a thing, that you can be good at one thing and bad at another. And people especially don’t like the notion that you can be better at the stylistic part of writing than the “making sense” part.
2 replies 3 retweets 15 likesShow this thread -
But look: my mom, who worked in academic publishing, could do a brilliant job editing a math manuscript for style without knowing calculus.
2 replies 1 retweet 14 likesShow this thread -
She also subconsciously imitated the accent of whoever she was speaking to. And picked up Romanian by listening to phone conversations. She had a freakishly good ear. I’m convinced this is a single phenomenon.
1 reply 0 retweets 11 likesShow this thread -
If I watch an episode of Peaky Blinders I’ll acquire an Irish accent by osmosis for the next hour. I subconsciously pastiche whichever writer I’ve been reading lately. I can’t *not* mirror conversational tone & emotion. One phenomenon.
1 reply 0 retweets 12 likesShow this thread -
I claim Scott Alexander has the same thing: writes compulsively, great at style pastiche, nobody believes him when he says he’s bad at math.
3 replies 0 retweets 19 likesShow this thread -
I think some people are legitimately better at style than sense or vice versa! They are not faking! Fluent pattern-matching != structural comprehension!
1 reply 0 retweets 17 likesShow this thread
Spectrum disorders may in part be explained by differences in attention. GPT-2 is a deeper model, but it is not a fully integrated one. Meaning is given by the relationships to a more or less unified model of the universe. I wonder if we need to go deeper, sparser or different.
-
-
Replying to @Plinz @s_r_constantin
Pattern matching and prediction has proved sufficient for creativity, but not understanding. I think reflection and *integration* (unification) are the sort of things needed for understanding. Yes, meaning is given by integrating ideas into a global model
1 reply 0 retweets 1 like -
I really doubt that you can squeeze the round peg of 'understanding' into the square hole of probability theory. Need some sort of extension to probability theory? For understanding, we're not *predicting*, we're *unifying* (integrating)
1 reply 0 retweets 1 like - 2 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.