300B output tokens, $20-30M in compute: Ethan Mollick says AI science will need far more compute
eldonredwards · x · 2026-09-09
- Ethan Mollick cites an AI-assisted math research effort whose 300 billion output tokens would cost a regular user an estimated $20-30 million, arguing that "whatever compute we have, it is increasingly likely that we are going to need more compute" as AI scales up scientific work.
- The quoted author also raises the question of how science can sustain this level of compute investment once AI companies deprioritize problems beyond Millennium Prize Problems.
More from AGI Musings
- ACM Computers and Society issue covers AI and democracy in Africa, debuts a Zine — ChinasaTOkolo · 2026-09-09
- Ex-pretraining researcher quits Anthropic, accusing OpenAI and Anthropic of racing recklessly to superintelligence — peterwildeford · 2026-09-09
- OpenAI's Millennium Prize breakthrough: compute scaling has ~4 OOMs left — sytelus · 2026-09-09
- Quant backgrounds have blind spots in the AI economy, says economist Daniel Rock — soumitrashukla9 · 2026-09-09
- Debate on ASI: why first-mover labs may not accumulate a decisive lead — zephyr_z9 · 2026-09-09
- AI job retraining ignores the wage collapse: a translator's story — PeterHndrsn · 2026-09-09