Tag
The paper introduces a label-free method for large language models to abstain when uncertain, using internal confidence signals, and shows it matches supervised abstention tuning in performance.
GoodfireAI shares research on using probes to improve LLM calibration and detect reasoning faithfulness, and a hackathon using their product Silico for targeted fine-tuning with parameter decomposition.
A new research paper introduces RLMF (Reinforcement Learning with Metacognitive Feedback), a two-stage approach that uses the model's own self-judgments to calibrate confidence and express uncertainty faithfully, achieving state-of-the-art calibration across diverse tasks while preserving accuracy and surpassing standard RL by up to 63%.
This paper introduces CAGE, a counterfactual graph-based method for calibrating multi-agent LLM systems, evaluating on benchmarks like TriviaQA and MMLU-Pro across various communication topologies. The method outperforms existing post-hoc and LLM-elicited calibration approaches.
This paper proposes Retrieval-Augmented Linguistic Calibration (RALC), a post-hoc pipeline for calibrating confidence signals in LLMs by modeling linguistic confidence as a distribution and using retrieval-augmented rewriting. It introduces Faithfulness Divergence metric and shows significant improvements across benchmarks.