Move 37 vs Zelnick: why datasets are a floor for creativity, not a ceiling
r0ck3t23 · x · 2026-09-04
Take-Two CEO Strauss Zelnick argues "a system built on data can never surprise you" and that datasets are backward-looking while creativity is forward-looking. The author dismantles this:
- Self-refuting: Zelnick admits creativity is informed by data — every book he read predates him, yet nobody says he can't have original thoughts
- Move 37: AlphaGo's 2016 move vs Lee Sedol, estimated at 1-in-10,000 for a human, sat unfound in 2,500 years of Go study; pros now call it beautiful
- AlphaFold: solved 50-year-old protein folding with structures never in training data — a Nobel Prize followed
- More: DeepMind broke the 1969 matrix multiplication record; in 2026 an unreleased OpenAI model disproved an Erdős conjecture from 1946, then produced ten more machine-verified results for $2,000 in tokens
Core point: treating a dataset as a ceiling is the error — it's examples with rules underneath, and holding the rules lets you build what examples never contained. The author concedes Zelnick's gaming points: asset creation is necessary but insufficient, and cheaper production only makes misses cheaper.
More from AGI Musings
- Prediction: the US builds the best AI products, China builds open-source AI — 0xsachi · 2026-09-04
- Dan Jeffries: no one ever regretted learning too much, even in the AI era — Dan_Jeffries1 · 2026-09-04
- Yoav Goldberg: every popular benchmark will be brute-forced, without real transfer — herbiebradley · 2026-09-04
- AGI doomer thread claims labor's economic value goes negative and the monetary system won't survive — Promptmethus · 2026-09-04
- Satirizing AI risk flip-flops: apologize, then ship a stronger model anyway — danfaggella · 2026-09-04
- Dan Jeffries: Never Regret Learning Anything — Even in the AI Era — Dan_Jeffries1 · 2026-09-04