Can LLMs Scale to AGI? | Hacker News Reader

Can LLMs Scale to AGI?

1 pointmr_rajat2 comments

is there any good argument in favor or against, I have read a lot but there is always the same argument like "they do not think", "they are next work predictors", "they are not biological". Which makes sense to some extent but does not explicitly tell why can't they scale to a completely intelligent system. Arguments like they are not "conscious" and "it's just dead math" doesn't make sense to me because we ourselves do not know what it means to be conscious and what actually is "thinking". And why is there need for biology or organics for intelligence? If you guys have some really good arguments, I'd love to hear.

View on HN

I mean, I do have an argument. AI, so giant, so much infrastructure, thing it couldnt do was replicate human thought, the unpredictability, the mess, it couldnt, maybe it will never. The spontaniety of the human mind, how you will be looking at cracks on a wall and come up with an idea for a painting, AI just never could have done that. Doesnt go to say AI is useless, but what Im saying is that these thinking models and idea generators are all worngly used and generate generic fluff that looks like all other gneric fluff, it gets nowhere. What AI does right now, even with so much infra, even with the best AI companies, Claude, etc, they excell at task automation, problem solving, but when it comes to building a "stance" which is consistent, and a "consistent personality", AI just cannot do it. Ai will just try its best to make you happy, it doesnt have a perspective, it gives you a general average of every available perspective. So no, whatever AGI is, intelligence, no I dont think AI is a concept that can scale to the same level as what we biologically are. Its not fundamentally possible, and we're already starting to see this.

You might want to check out Yann LeCun’s arguments—they’re pretty interesting. TL;DR: he sees LLMs as a dead end and backs that up with concrete examples.

https://www.youtube.com/watch?v=vJKC31YpA8c

Reply on news.ycombinator.com