@scaling01, who runs the LisanBench LLM reasoning benchmark, shares timeline showing GPT-4o below 10 percent on AIME 2024, o3-preview saturation by December 2024, and first Millennium Prize solution projected for 2028
Andrew Curran extends the timeline to a 2030 trumpets entry.
@scaling01 2030: The distant sound of trumpets
Lisan al Gaib@scaling01May 13, 2024: GPT-4o is released and scores below 10% on AIME 2024 (US math olympiad entry questions) Dec 20, 2024: o3-preview saturates AIME 2024 Jul 21, 2025: Gemini DeepThink scores IMO gold May 20, 2026: internal GPT models makes Erdös unit-distance problem breakthrough 2027 ... 2028 ... 2029 ... 2030 ... 2031 ... 2032 ... I don't think this prediction is very far fetched. I don't know about the "all but one" part. But I think the first Millennium Prize problem falls in 2028.
#980Lisan al Gaib@SCALING01I also think many people are completely underestimating the amount of compute we will have by 2032 and the willingness to spend a gigantic amount of money on solving these really difficult, almost impossible problems.
If you think a billion tokens is a lot, think several OOMs bigger.
Lisan al Gaib@scaling01May 13, 2024: GPT-4o is released and scores below 10% on AIME 2024 (US math olympiad entry questions) Dec 20, 2024: o3-preview saturates AIME 2024 Jul 21, 2025: Gemini DeepThink scores IMO gold May 20, 2026: internal GPT models makes Erdös unit-distance problem breakthrough 2027 ... 2028 ... 2029 ... 2030 ... 2031 ... 2032 ... I don't think this prediction is very far fetched. I don't know about the "all but one" part. But I think the first Millennium Prize problem falls in 2028.
and I think it's going to be the Birch-Swinnerton-Dyer conjecture or the Hodge conjecture that's first going to get solved
Lisan al Gaib@scaling01May 13, 2024: GPT-4o is released and scores below 10% on AIME 2024 (US math olympiad entry questions) Dec 20, 2024: o3-preview saturates AIME 2024 Jul 21, 2025: Gemini DeepThink scores IMO gold May 20, 2026: internal GPT models makes Erdös unit-distance problem breakthrough 2027 ... 2028 ... 2029 ... 2030 ... 2031 ... 2032 ... I don't think this prediction is very far fetched. I don't know about the "all but one" part. But I think the first Millennium Prize problem falls in 2028.
I would extend my win condition a bit. Because I think the first Millennium Prize will be an AI assisted proof. So human + AI.
A pure AI only approach I see more towards 2029-2030
Lisan al Gaib@scaling01May 13, 2024: GPT-4o is released and scores below 10% on AIME 2024 (US math olympiad entry questions) Dec 20, 2024: o3-preview saturates AIME 2024 Jul 21, 2025: Gemini DeepThink scores IMO gold May 20, 2026: internal GPT models makes Erdös unit-distance problem breakthrough 2027 ... 2028 ... 2029 ... 2030 ... 2031 ... 2032 ... I don't think this prediction is very far fetched. I don't know about the "all but one" part. But I think the first Millennium Prize problem falls in 2028.
It seems likely to me that this hybrid proof will happen with an internal 100T+ reasoning model
similar to how Mythos is used right now by cyber experts
Lisan al Gaib@scaling01I would extend my win condition a bit. Because I think the first Millennium Prize will be an AI assisted proof. So human + AI. A pure AI only approach I see more towards 2029-2030
that is, late 2028
that's when I expect 100T models
Lisan al Gaib@scaling01It seems likely to me that this hybrid proof will happen with an internal 100T+ reasoning model similar to how Mythos is used right now by cyber experts