The problem here is that, if you will excuse the reference, "attention isn't all you need"[0].
There's a lot which AI is still bad at.
Fortunately, because if it was good at everything, it wouldn't just be mathematicians and software developers (and artists, and translators…) having an "is my career dead?" crisis.
Unfortunately, the areas AI is still bad at get pointed to by a subset of the population who are unwilling to accept there is any risk to themselves, either in business ("my job can't be taken by an AI") or in a more existential sense ("AI can't destroy the world, it's just software").
Being good at maths doesn't really change the latter group. I was kinda hoping the various hacking incidents on https://www.felonybench.com would convince people that it's a cyber threat, but there's still loads who insist this is all a PR scam for the IPO, so… yeah. Dunno what'll convince them.
Economic metrics and data versus very specific domain superiority. Decades of math done quickly is like BOINC or distributed.net computing projects, useful brute force versus general intelligence.
LLMs can be very useful (formal proof brute forcing, software factories, highly optimized pattern matching within and ancross domains and their knowledge graphs) and yet not world changing. Another 6-12 months should show us more, maybe 18 months.
It has our attention.
The problem here is that, if you will excuse the reference, "attention isn't all you need"[0].
There's a lot which AI is still bad at.
Fortunately, because if it was good at everything, it wouldn't just be mathematicians and software developers (and artists, and translators…) having an "is my career dead?" crisis.
Unfortunately, the areas AI is still bad at get pointed to by a subset of the population who are unwilling to accept there is any risk to themselves, either in business ("my job can't be taken by an AI") or in a more existential sense ("AI can't destroy the world, it's just software").
Being good at maths doesn't really change the latter group. I was kinda hoping the various hacking incidents on https://www.felonybench.com would convince people that it's a cyber threat, but there's still loads who insist this is all a PR scam for the IPO, so… yeah. Dunno what'll convince them.
[0] https://en.wikipedia.org/wiki/Attention_Is_All_You_Need
Economic metrics and data versus very specific domain superiority. Decades of math done quickly is like BOINC or distributed.net computing projects, useful brute force versus general intelligence.
LLMs can be very useful (formal proof brute forcing, software factories, highly optimized pattern matching within and ancross domains and their knowledge graphs) and yet not world changing. Another 6-12 months should show us more, maybe 18 months.