alt Hacker News> Just because it's not that hard to reach a high-level understanding of the transformer pipeline doesn't mean we understand how these systems function
Mumbo jumbo magical thinking.
They perform so well because they are trained on probabilistic token matching.
Where they perform terribly, e.g math, reasoning, they are delegating to other approaches, and that's how you get the illusion that there is actually something there. But it's not. Faking intelligence is not intelligence. It's just text generation.
> In that sense, yeah you could say they are a bit "magical"
Nobody but the most unhinged hype pushers are calling it "magical". The LLM can never ever be AGI. Guessing the next word is not intelligence.
> there can be no form of world model that they are developing
Kind of impossible to form a world model if your foundation is probabilistic token guessing which is what LLMs are. LLMs are a dead end in achieving "intelligence", something novel as an approach needs to be discovered (or not) to go into the intelligence direction. But hey, at least we can generate text fast now!
> LLMs are a dead end in achieving "intelligence"
There is no evidence to indicate this is the case. To the contrary, all evidence we have points to these models, over time, being able to perform a wider range of tasks at a higher rate of success. Whether it's GPQA, ARC-AGI or tool usage.
> they are delegating to other approaches > Faking intelligence is not intelligence. It's just text generation.
It seems like you know something about what intelligence actually is that you're not sharing. If it walks, talks and quacks like a duck, I have to assume it's a duck[1]. Though, maybe it quacks a bit weird.
[1] https://en.wikipedia.org/wiki/Solipsism