To all the people that are confused by "skeptics": Yes, LLMs are powerful tools
Yes, LLMs can be useful
I think "skeptic" is the wrong word for most of us. There's certainly people that will just deny that LLMs and ML can do anything useful. But I don't think that's the majority of the "skeptic" crowd.I'm an ML[0] researcher. Of course I love these machines! And you know what? I really fucking want to build some god damn AGI. Maybe all you hype people are all right and LLMs will get us there. If you're right, you got all the money you could ask for and then several times more. You got some of the biggest companies in the world dumping billions in, competing, with huge infrastructures, and dumping money into a few dozen startups who are trying to do the same. If scale is all you need, then you're going to get there even without another 7 trillion dollars. I'm not saying "Don't fund LLM research" nor am I saying "Don't fund scaling research". Please do fund those things. But be honest, money isn't your bottleneck. Hell, what if you're right but other ways are faster?! We miss those opportunities and lose the race? For what?! If more efficient opportunities are out there (and I really do believe there are), who do you think is going to find it? It's literally betting on ITER[1] and giving essentially nothing to other fusion attempts. Except ITER is a international effort and we're a hell of a lot more confident it'll work. But what if you're wrong? Why stop people like me from looking at other avenues? Why put all your eggs in one basket? Why dismiss us when we point out issues? We're engineers here, we fucking love to point out issues! IT IS OUR JOB! Because you must find issues if you're going to fix them.
We could have hedged our bets a little and pivot fast, hopefully seamlessly enough that it doesn't even look like a hiccup.
Nobody in this side of this is asking for half as much as what the next hip new LLM startup is. I don't know if it is even a penny on the dollar! Are you really gonna bet that the golden goose is always going to be laying golden eggs? Are you really going to make a situation where if it stops you just go broke? IDK, it just all seems needlessly risky to me. Maybe I just don't get it because I'm not a high stakes gambler. [0] Funny... only a few years ago the litmus test was "If someone says 'AI' they're selling you something. If they say 'ML' they're probably not" [1] https://en.wikipedia.org/wiki/ITER |
What happens to the majority of the planet's biomass when we get there? If it's a new paradigm that does the trick. It might sneak up on us like the GPT-2 capabilities.