Sebastien Bubeck claims GPT-5.5 can replicate a unit distance problem solution, fueling debate over frontier models making novel scientific discoveries
Critics argue reproducing known solutions does not equal novel discovery.
@iruletheworldmo It depends on what you mean. To some extent, TRUE breakthroughs are very hard to identify for humans too. How long did it take before the world understood that the growing neural nets of the 2000s were a fundamental breakthrough?
🍓🍓🍓@iruletheworldmo@SebastienBubeck how far away are we from models being able to discern what a breakthrough looks like?
@SebastienBubeck how far away are we from models being able to discern what a breakthrough looks like?
Sebastien Bubeck@SebastienBubeckWe now know that with an appropriate harness both Mythos and GPT-5.5 can reproduce what our internal model did in one-shot for the unit distance problem. Clearly there is an insane overhang of capabilities with this generation of models, and no ceiling in sight for what scientific advances they can bring. You can go and try to discover new things with 5.5 right now!
true but at the moment it feels like models are throwing ideas at the wall and then humans need to spend a long time verifying.
seeing the same trend in cyber.
so how long before models have better taste, oversight, research direction than us.
it feels right now like we have to steer initially in the right direction a lot and then verify a lot at the end. which in the short turn i’d guess causes more work than we can handle.
Sebastien Bubeck@SebastienBubeck@iruletheworldmo It depends on what you mean. To some extent, TRUE breakthroughs are very hard to identify for humans too. How long did it take before the world understood that the growing neural nets of the 2000s were a fundamental breakthrough?
@SebastienBubeck i take your point though maybe i’m holding them to a higher standard than us
Sebastien Bubeck@SebastienBubeck@iruletheworldmo It depends on what you mean. To some extent, TRUE breakthroughs are very hard to identify for humans too. How long did it take before the world understood that the growing neural nets of the 2000s were a fundamental breakthrough?