Mx. Chara Aznable (they/them) of Pnictogen on Nostr: All the U.S. #media and #journalism on the "generative AI" craze refuses to treat ...

All the U.S. #media and #journalism on the "generative AI" craze refuses to treat with this basic problem: the thing that Anthropic and OpenAI and all the other genAI vendors are selling can scarcely be called *intelligent* at all, if we define "intelligent" in terms of the ability to make abstractions and generalizations, and then reason on the basis of those generalizations.

If you think "intelligence" is merely a matter of raw data shoved into the system, as if "intelligence" in the thinking sense were the same as "intelligence" in the spywork sense (i.e. "intel", any info gleaned through spying) then I suppose yes, LLMs must seem like they must necessarily be more brilliant than any human being is capable of achieving. But it's how a being *uses* the information they've learned that matters, and in that regard the #LLM is a bad joke. No matter how big they get, no matter how much hardware and data are crammed into LLMs, they get no smarter. They do not really learn, except in the sense that they have more tokens (i.e. individual chunks of text) to play with.

They're all still confabulating away, emitting mixed-up gibberish that sometimes accidentally resembles meaningful and intelligent communication.

(cont'd)