Gacha Decoding diversifies LLM outputs with 11x better sample efficiency
natolambert · x · 2026-10-03
Researchers introduce Gacha Decoding, a new method for diversifying LLM responses that beats prior work by over 11x in sample efficiency. Instead of relying on token entropy, it leverages the model's instruction-following ability plus an external RNG — reroll the LM like a gacha game to get varied ideas, synthetic data, or RL environments.
More from Research
- AC2: actor-critic with action chunking enables partial rollouts, trains faster than GRPO — ZeYanjie · 2026-10-03
- KAIST's World Observer gives world models movable panoramic eyes to track unseen regions — Scobleizer · 2026-10-03
- Ofir Press defends new bug-finding benchmark: training the behavior is fine, test-set contamination is not — OfirPress · 2026-10-03
- CAS Five-Year Plan gives AI for Science its own chapter, setting up a US-vs-China metascience bet — teortaxesTex · 2026-10-03
- Draft standard for autonomous labs emerges as AI-driven science heats up — Afinetheorem · 2026-10-03
- Papers with 'agent' in the title get cited more: AI lit search reads titles too — rajammanabrolu · 2026-10-03