This mathematician recently been coming up in my YouTube feed (thanks discovery Algo!) and he gives a great explanation why AI can’t solve complex mathematical problems that been around for decades.
He discusses how AI helps mathematicians use an obscure method to a problem and once it’s solved it gets massive hype but the actual hit rate is about 1 to 2%!!
We have a long way to go to achieve AGI
Fork lifts can lift more than a dozen men and do it all day every day. LLMs can write prose and even code very effectively with the right guidance. AGI to me is a silly hype thing pitched by dudes trying to scam people out of money more than anything.
The reality is more boring but also probably better. We have computers that are amazing a math. Have for a long time. People that just want to score points in a dumb AI sucks reaction game love to point out what LLMs are bad at.
Instead of being reactionary we should be wide eyed and skeptical. LLMs are quite amazing at what they do well but they are no where near some SciFi intellect with a will to do things.
We are at the same time being manipulated by fear to give the state more power and these companies more money. As bitcoin shows us, always follow the incentives.
I agree that the capabilities are overhyped, but I feel like the way you described it undersells it too. I think it's a genuinely new and revolutionary capability.
Ok, but only because all my examples are so old to us. When hydrolics and internal combustion engines were first combined it was also revolutionary. The age of machines had arrived. This was tech. Now it's nothing to us.
Its funny, I'm not trying to undersell anything just push back a bit. Which usually means I get accused by both extremes of having a dumb take.
Also, I was responding to the AGI part. That's massively hype. And we can see this as we have been 3/months away for years now.
Even in the field of mathematics?
Harder for me to comment on that, but even if AI isn't able to make progress on frontier questions, they should streamline a lot of the other work (even stuff like finding and synthesizing other peoples' research), that it should significantly accelerate math development, IMO.
By humans which is the point made by Mr. Tao. AI has a use case it’s not the AGI we are all hoping for
Agreed.... still very much revolutionary though, IMO. I don't see a slowdown to AI demand. Question for stocks is whether the growth is priced in or not, and whether it's overly priced in or by not enough.
The stock market is way overvalued but that’s just the population trying to protect its purchasing power.
I bet you at least 50% of AI investors don’t care about the tech they just want the growth.
Some AI just struggles with math, period. Copilot can't even add properly.
Indeed. We should realize by now that making a machine that can do one or even multiple things better than humans is one thing. Been done many many times. Making a machine that be better at all things than a human is a long way off.
Cool, I actually shared a Quanta Magazine piece about this with Tao today!