DART: ambiguity steering vectors improve LLM confidence calibration across 7 models
dhadfieldmenell · x · 2026-11-11
LLMs are highly overconfident even when they don't know something. New work led by Themistoklis Nikolas proposes DART, which derives ambiguity steering vectors via interpretability and uses them to guide fine-tuning — improving confidence calibration across 7 models while preserving accuracy. The vectors can also encode these improvements directly into model weights.
Related event: DART uses ambiguity steering vectors to curb LLM overconfidence(2 posts)→
More from Research
- MacroHands debuts: 3D hand tracking from egocentric video, stereo variant cuts depth error to 13.1mm — eliebakouch · 2026-11-11
- Yann LeCun: 'If you are interested in human-level AI, don't work on LLMs' — StasBekman · 2026-11-11
- FlashAttention co-creator Dan Fu on kernels, inference economics, and owning vs leasing GPUs — realDanFu · 2026-11-11
- OpenTac debuts at CoRL as researchers call tactile the missing half of robot data — anand_bhattad · 2026-11-11
- Geodesiclab scores big improvement on antibody structure prediction, the hardest part of the problem — QuanquanGu · 2026-11-11
- AI drug discovery hits its HBM moment as the bottleneck shifts to the wet lab — QuanquanGu · 2026-11-11