Verifying ML research may cost 10^4-10^9x more than verifying math
norvid_studies · x · 2026-11-22
norvidstudies agrees with the broad thrust of 1a3orn's takeoff analysis but flags the missing link: nobody has quantified how much more compute verifying an ML research result takes versus verifying a math proof — he guesses on the order of 10^4 to 10^9x, and that gap effectively becomes your training cost. He calls for a proper quantitative analysis of this step.
Related event: Cost of Verifying Largest Lean Projections Revealed(2 posts)→
More from AGI Musings
- AI bubble panic is orchestrated, says dev citing model-run analysis: Oracle 0.8% default risk, Nvidia <8% China exposure — sytelus · 2026-11-22
- Math was rediscovered for centuries; LLMs finally made theorems searchable — konstmish · 2026-11-22
- Will everyone publish a 'values constitution' for their AI agents? — round · 2026-11-22
- The underrated detail in 'Claude solved it': Fields-medal-caliber minds with AI tools — avlok · 2026-11-22
- Rob Miles to Anthropic: 'You're good people, but you might kill us all' — zetalyrae · 2026-11-22
- McLuhan's 1970 note resurfaces: obsolescence is the matrix of innovation — round · 2026-11-22