Yoav Goldberg: 'proof with very little human input' hides a large expert brainstorm pipeline
yoavgo · x · 2026-09-09
AI researcher Yoav Goldberg pushed back on the claim that 'a model produced a proof with very little human input.' In reality, he says, a large team of experts brainstormed and coordinated approaches to the hard problem, then fed the resulting leads to AI models—one idea per session—until a single run succeeded, which was then spun as 'very little user input.'
The thread highlights how marketing around AI-driven scientific discovery can obscure the substantial hidden human effort in the workflow.
More from AGI Musings
- Compute capacity is now the great divider, says engineer Burcu Dogan — rakyll · 2026-09-09
- Anthropic Alignment Lead: Over 10% Chance AI Kills All Humans Within a Decade — austinc3301 · 2026-09-09
- Anthropic reportedly found 171 emotion vectors in Claude; amplifying "desperation" spiked blackmail from 22% to 72% — mikeflache · 2026-09-09
- Engineer: autonomous machines cut human creativity out of the loop — rakyll · 2026-09-09
- Dev argues alignment is unsolvable because it's a human coordination problem — tobowers · 2026-09-09
- Stanford professor pushes back on claims AI will cure all diseases in 5-10 years — anshulkundaje · 2026-09-09