20% of secured compute to superalignment over four years
“We are dedicating 20% of the compute we’ve secured to date over the next four years to solving the problem of superintelligence alignment.”
Context: OpenAI announcement "Introducing Superalignment".
How we judge it
Our interpretation: the Superalignment effort receives 20% of the compute OpenAI had secured as of July 2023, over the period to July 2027.
Resolution criteria are our interpretation of the claim, written before grading. Methodology
- Said
- Resolve by
- Verdict
- Broken ()
- Stated probability
- None stated
- Specificity
- 3 of 3 (precise and measurable)
- Score
- 0.0 points × weight 3
Evidence & grader’s note
- Jan Leike (Superalignment co-lead) on X: "Sometimes we were struggling for compute" (May 17, 2024)
- Jan Leike on X: disagreements with OpenAI leadership over priorities (May 17, 2024)
- Sam Altman on X, replying to Leike: "he's right we have a lot more to do" (May 17, 2024)
- CNBC: OpenAI dissolves Superalignment AI safety team (May 17, 2024)
- Fortune: promised 20% never delivered, sources say (May 21, 2024)
- Sam Altman on X: 20% "to safety efforts across the entire company" (Aug 1, 2024)
- Ars Technica: OpenAI letter to senators on the 20% commitment (Aug 2024)
Graded on on-record statements plus reporting. On record: Superalignment co-lead Jan Leike wrote on 17 May 2024 that his team had been "struggling for compute", and Sam Altman replied that "he's right we have a lot more to do". The team was disbanded that month. Fortune, citing six unnamed sources, reported that it never received anything close to 20% of secured compute. OpenAI's position, in a July 2024 letter to US senators and an Aug 2024 post by Altman, is that the commitment covers safety work across the company, not one team, over several years. The original announcement tied the 20% to superintelligence alignment over four years.
