Collective Bias Mitigation: Multi-LLM Committee Cuts Age Bias Score From 0.25 to 0.10
NUS-IDS · hf · 2026-10-05
NUS-IDS researchers propose Collective Bias Mitigation (CBM), a framework that reduces LLM bias by routing and organizing multiple models rather than relying on single-model self-debiasing. In a top-7 setting, the Committee topology lowers age bias score from 0.25 to 0.10, substantially outperforming standalone baselines. The work systematically explores model selection and collaboration topologies (Debating vs. Committee), with Committee balancing mitigation effectiveness and inference cost.
More from Research
- First large-scale 3B/8B continuous diffusion LMs match pass@1 and beat pass@k vs masked dLMs — ArashVahdat · 2026-10-06
- AggAgent at COLM: Treats Parallel Agent Trajectories as an Environment for Long-Horizon Tasks — xiye_nlp · 2026-10-06
- New paper: LLM verbalized probabilities directly track internal uncertainty, not just training data — sineadwilliamso · 2026-10-06
- Nature Biotech paper: "single-cell models don't beat baselines" was a measurement problem — BoWang87 · 2026-10-06
- 250+ AI experts broadly agree: international body with pre-release power best cuts risk — robseamans · 2026-10-06
- Negative result: state-cosine adaptive depth stopping rule hurts both loss and passes — BlackHC · 2026-10-06