upvote
I don't see anybody claiming that solving the Riemann hypothesis would be proof that current models are superintelligent. Reminder that the definition of the term is "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest". I do sometimes see people arguing that current models are AGI, but I don't think even this is true - current models are slightly superhuman in some narrow domains like math, human-level at a bunch of them, and still pretty stupid in some ways, most notably agency.

The reason people worry about superintelligence is that one of the "domains of interest" is AI research, and once you have an AI that's as good at AI research as your human researchers, your AI progress shoots up, probably exponentially. So the delay between automated AI research and superintelligence might be tiny.

reply
You can believe what you want I probably agree with your nuanced view. But I would invite you to control+F this thread for “superintelligence” there are plenty of aggrandizing statements about what fools everyone are for being dubious about this being superintelligence.

(e.g)[https://news.ycombinator.com/reply?id=50575797&goto=item%3Fi...]

People are just tired of listening to these statements and being called idiots constantly with no evidence other than other press releases and disingenuous framing getting echoed constantly in volume across corporate media and all social media that allows LLMs.

reply
I don’t think like that, the most important question is: could he have solved the problem without AI assistance, and in the same time frame? I’d guess no.
reply
A few years ago if someone had said “we used an HPC cluster and Pandas to build some tools that in turn solved this problem” no-one would’ve batted an eyelid and no-one would’ve been making grandiose statements about the end of the world due to HPC computers and the power of Dataframes. Personally I think If someone had had a few hundred billion dollars of investment pumped into a bunch of mathemeticians to solve this problem then yes they probably would have solved it too and maybe that might have raised some eyebrows. Everyone of these conversations just feels like an extremely highly charged argument about Vim vs VSCode.
reply