alt Hacker NewsA lot of C-suite people seem to have an idea that if they just throw enough compute at LLMs that AGI will eventually emerge, even though it's pretty clear at this point that LLMs are never going to lead to general intelligence. In their view it makes sense to invest massive amounts of capital because it's like a lottery ticket to being the future AGI company that dominates the world.
I recall Zuckerberg saying something about how there were early signs of AI "improving itself." I don't know what he was talking about but if he really believes that's true and that we're at the bottom of an exponential curve then Meta's rabid hiring and datacenter buildout makes sense.
Replies
Its insane really, anyone who has worked with LLMs for a bit and has an idea of how they work shouldn't think its going to lead to "AGI".
Hopefully some big players, like FB bankrupt themselves.
> I don't know what he was talking about
There's a bunch of ways AI is improving itself, depending on how you want to interpret that. But it's been true since the start.
1. AI is used to train AI. RLHF uses this, curriculum learning is full of it, video model training pipelines are overflowing with it. AI gets used in pipelines to clean and upgrade training data a lot.
2. There are experimental AI agents that can patch their own code and explore a tree of possibilities to boost their own performance. However, at the moment they tap out after getting about as good as open source agents, but before they're as good as proprietary agents. There isn't exponential growth. There might be if you throw enough compute at it, but this tactic is very compute hungry. At current prices it's cheaper to pay an AI expert to implement your agent than use this.
I don't get it, I really don't.
Even assuming a company gets to AGI first this doesn't mean another one will follow.
Suppose that FooAI gets to it first: - competitors may get there too in a different or more efficient way - Some FooAI staff can leave and found their own company - Some FooAI staff can join a competitor - FooAI "secret sauce" can be figured out, or simply stolen, by a competitor
At the end of the day, it really doesn't matter, the equation AI === commodity just does not change.
There is no way to make money by going into this never ending frontier model war, price of training keeps getting higher and higher, but your competitors few months later can achieve your own results for a fraction of your $.
I don't know if AGI will emerge from LLM, but I'm always reminded of the Chinese room thought experiment. With billions thrown at the idea it will certainly be the ultimate answer as to whether true understanding can emerge from a large enough dictionary.
In early 2023, I remember someone breathlessly explaining that there are signs that LLMs that are seemingly good at chess/checkers moves may have a rudimentary model of the board within them, somehow magically encoded into the model weights through the training. I was stupid enough to briefly entertain the possibility until I actually bothered to develop a high level understanding of the transformer architecture. It's surprising how much mysticism this field seems to attract. Perhaps it being a non-deterministic, linguistically invoked black box, triggers the same internal impulses that draw some people to magic and spellcasting.