GeePawHillMay 19

It's too bad that LLM grifters sought to shine their reputation by calling it "AI".

It didn't shine their reputation.

It ruined the world's opinion of the fascinating & legitimate science -- and remarkably creative uses -- of actual artificial intelligence.

Show threadSigma
@GeePawHill Hmm… I’m not sure that’s true. I mean, probably depends on what you define as "actual artificial intelligence" but traditionally natural language processing is part of the field of artificial intelligence. And LLMs essentially solved natural language processing. So they are arguably a huge leap in AI research. And related fields like image recognition, speech recognition, … also much big steps forward because of LLMs - or more specifically the transformer architecture.

And I personally also think that agentic AI using LLMs is super fascinating from a research perspective. But is that enough to cancel out all the bad LLMs are causing? Probably not.
May 20 at 8:44amWeb
Show threadElse, Someone3d ago

@sigmasternchen @GeePawHill

> And LLMs essentially solved natural language processing.
> So they are arguably a huge leap in AI research.
> ...big steps forward because of LLMs - ...transformer...

Naaaah call me a hater, but IMO all 3 statements are somewhat false.

Judging by the public picture, it feels like there has hardly been any progress in language modeling in the past 5-6 years (research halted by chatgpt), and only modest progress in the past two decades (incl. transformers).

Show threadElse, Someone3d ago
@sigmasternchen @GeePawHill unless openai/anthropic do something wildly different internally, modern LLMs still remain NeRFs of language modeling
Show threadElse, Someone3d ago
@sigmasternchen @GeePawHill for the impact of transformers specifically... I'd need to collect my receipts but I very much suspect it must be the same story as Alyosha Efros' "Victory of Data", i.e. that we're still in the mode of "running ANN over increasingly large datasets"