SPoT results: +17.7pp safety across 5 benchmarks, +4.7pp math accuracy over RL
polkirichenko · x · 2026-11-12
Part 3 of the SPoT thread presents headline results:
- Safety: +17.7pp generation safety across 5 benchmarks over the base model, plus another +3.6pp from self-selection
- Reasoning: +4.7pp accuracy across 4 math benchmarks (MATH500, GSM8K, AMC23, OlympiadBench) over standard RL post-training, plus +6.3pp from self-selection
- Gains hold both in raw generations and when self-judgment picks the best rollout; weighted majority voting with post-thinking reaches 65.3 on MATH500, 90.1 on GSM8K, 40.1 on GPQA-D
Related event: Meta's SPoT: self-improving post-thinking training(4 posts)→
More from Research
- As models get stronger, the next trainable tasks look increasingly like alignment itself — herbiebradley · 2026-11-12
- RoboCap-10k: 10,000 hours of open egocentric robot data across 21 industries released — micoolcho · 2026-11-12
- Talent vs. luck model reproduced: equal funding is ~9x more efficient than funding winners — ChenhaoTan · 2026-11-12
- Stanford HAI dashboards show employment in AI-exposed jobs down over 4% this year — StanfordHAI · 2026-11-12
- Shun-ichi Amari on how 'do as you like' freedom birthed the math behind AI's 60 years — FrnkNlsn · 2026-11-12
- Formal verification proves a 3,000-step Sokoban optimum, with a bounty on beating it — banteg · 2026-11-12