A followup thought: even nuclear war probably wouldn’t prevent hard asi by 2030. It would still be incredibly high leverage for the crazy shack scientists, of which there would be plenty left, nuclear war isn’t as deadly as people think before learning about it in detail.
soft asi, semi-general planners that can trounce humans at very difficult and complex well defined tasks, this year or next. hard asi, planners that can trounce humans at absolutely everything in an efficient package, less certain because soft ASI might slow things down rather than speed things up if safety goes well, but by 2028 it seems all but guaranteed to me. I haven’t convinced other ppl so they buy it back up when I bet this, similar to the flood of people disagree-voting here. still gonna happen though.
soft asi, semi-general planners that can trounce humans at very difficult and complex well defined tasks, this year or next [...] soft ASI might slow things down rather than speed things up if safety goes well
I think most people expressing predictions about IOI and IMO expect some self-play tricks with proof generation for verification of formal statements, which don’t translate into AGI in a singularity-relevant sense, the same as MCTS for Go and chess doesn’t. So while I agree that IOI and IMO bets should be AGI-equivalent (if I’m reading the “slow things down” point correctly), I think the prediction aggregators don’t have that claim baked in for such questions.
A followup thought: even nuclear war probably wouldn’t prevent hard asi by 2030. It would still be incredibly high leverage for the crazy shack scientists, of which there would be plenty left, nuclear war isn’t as deadly as people think before learning about it in detail.
What year do you put the arrival of ASI without nuclear war?
soft asi, semi-general planners that can trounce humans at very difficult and complex well defined tasks, this year or next. hard asi, planners that can trounce humans at absolutely everything in an efficient package, less certain because soft ASI might slow things down rather than speed things up if safety goes well, but by 2028 it seems all but guaranteed to me. I haven’t convinced other ppl so they buy it back up when I bet this, similar to the flood of people disagree-voting here. still gonna happen though.
I think most people expressing predictions about IOI and IMO expect some self-play tricks with proof generation for verification of formal statements, which don’t translate into AGI in a singularity-relevant sense, the same as MCTS for Go and chess doesn’t. So while I agree that IOI and IMO bets should be AGI-equivalent (if I’m reading the “slow things down” point correctly), I think the prediction aggregators don’t have that claim baked in for such questions.