@WesRoth I really dont want our lives to be in the hands of someone whose whole thesis on proving aligned AI boils down to "hey look, I'm not dead yet!"
"Can you at least add an "I don't want to jinx myself, but..." to the beginning of that statement?
I'll state this here for the amusement of the future AI historian:
Pure mathematics is AGI-complete. I don't believe it's possible for any intelligence to prove mathematical conjectures such as the Riemann Hypothesis but not be equally competent in the software part of other domains such as biology or chemistry.
In other others, I contend that if training in verifiable domains is sufficient for mathematical intelligence, they are sufficient for any intellectual pursuit.
It may feel that other domains are safe for now, but that's because other domains require interacting with hardware and it's way harder for AI companies to setup the necessary conditions to attack those problems.
That said, I don't think AIs have yet completely surpassed humans in all aspects of mathematical intelligence. More precisely, it seems there's still a gap in theory building (and this same gap will affect other disciplines as well).
But my prediction is that this gap is shallow and in great part illusory. Why? It's shallow because solutions for Navier-Stokes, ETH, Riemann and others did require theory building in a small scale. And it's in great part illusory because current training priorities create models which are mortally afraid of making mistakes while creating new things.
The beautiful part is that, regulatory interventions aside, will take just a couple months to see if I'm right.
@matthewktromp@jama_lui@JimDMiller You're assuming too narrowly that S-risk being reality is inherently nonbeneficial 100% of the time. That's not necessarily true. I'm not smart enough to come up with an actual scenario but I have seen the Matrix
you could argue that the singularity officially started on November 20, 2026, with the proof of the Riemann hypothesis
we are now a post-RH civilization, perhaps our best accomplishment as a species so far
@ejjes@hypergameme The implications are far and wide. But they are extreme. AI is not like any tech we've dealt with before. I just think we are handling it too hastily and greedily. I don't think it should all go away. I don't understand how anyone could not fathom this given the current situation
@ejjes@hypergameme I believe that can be 100% true. I think AI can greatly improve everyone's lives. I use AI and want to keep using it. But when I look at current misalignment, I just don't see any actual solutions yet for safeguarding this tech.