2024 tweet predicting 5T-param model solving Riemann hypothesis by 2026 resurfaces

scaling01 · x · 2026-11-21

scaling01 resurfaced his own December 2024 tweet: back then the most-used models were all under 500B params, but he predicted GB200 NVL72 would enable 10x larger models within 1-2 years — half-joking that a 5T/10T param model with test-time compute would solve the Riemann hypothesis in 2026. In a follow-up he clarified he was joking, expecting most Millennium Prize Problems to fall in 2027/2028.

Related event: Old Tweet Predicting AI Would Crack the Riemann Hypothesis Goes Viral(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →