REVIEW 9 cited by
Probabilities of Chat LLMs Are Miscalibrated but Still Predict Correctness on Multiple-Choice Q&A
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
We study 15 large language models (LLMs) fine-tuned for chat and find that their maximum softmax probabilities (MSPs) are consistently miscalibrated on multiple-choice Q&A. However, those MSPs might still encode useful uncertainty information. Specifically, we hypothesized that wrong answers would be associated with smaller MSPs compared to correct answers. Via rigorous statistical testing, we show that this hypothesis holds for models which perform well on the underlying Q&A task. We also find a strong direction correlation between Q&A accuracy and MSP correctness prediction, while finding no correlation between Q&A accuracy and calibration error. This suggests that within the current fine-tuning paradigm, we can expect correctness prediction but not calibration to improve as LLM capabilities progress. To demonstrate the utility of correctness prediction, we show that when models have the option to abstain, performance can be improved by selectively abstaining based on the MSP of the initial model response, using only a small amount of labeled data to choose the MSP threshold.
Forward citations
Cited by 9 Pith papers
-
Instance-Adaptive Online Multicalibration
A single online multicalibration algorithm adaptively refines a dyadic grid and achieves instance-dependent rates: O(T^{2/3}) worst-case, O(sqrt T) for marginal stochastic data, and O(sqrt(JT)) for J-piecewise station...
-
Task-Aware Calibration: Provably Optimal Decoding in LLMs
Task calibration aligns LLM distributions in latent task spaces to make MBR decoding provably optimal and improve generation quality.
-
Instance-Adaptive Online Multicalibration
A single algorithm for online multicalibration achieves instance-adaptive rates by dynamically refining a dyadic prediction grid, recovering the worst-case Õ(T^{2/3}) bound and improving to Õ(√T) in marginal stochasti...
-
PromptNCE: Pointwise Mutual Information Predictions Using Only LLMs and Contrastive Estimation Prompts
PromptNCE frames LLM conditional probability estimation as contrastive prompting augmented with an OTHER category, recovering true P(y|x) and achieving up to 0.82 Spearman correlation with human-derived PMI on three datasets.
-
Causal Evidence that Language Models use Confidence to Drive Behavior
Language models deploy multidimensional internal confidence representations and threshold-based policies to control abstention behavior, with causal support from activation steering experiments.
-
Calibrating Model-Based Evaluation Metrics for Summarization
A reference-free proxy scoring framework combined with GIRB calibration produces better-aligned evaluation metrics for summarization and outperforms baselines across seven datasets.
-
Rethinking Uncertainty Estimation in LLMs: A Principled Single-Sequence Measure
Negative log-likelihood of the greedy-decoded most likely sequence (G-NLL) is a principled single-sequence uncertainty measure for LLMs that achieves state-of-the-art results.
-
Enhancing Trust in Large Language Models via Uncertainty-Calibrated Fine-Tuning
Uncertainty-aware fine-tuning with a decision-theory-based loss produces better-calibrated uncertainty estimates than standard training on free-form QA tasks.
-
ConSteer-RL: Steering Reasoning Capabilities in Large Language Models via Confidence-Aware Reinforcement Learning
ConSteer-RL adds a confidence-aware reward derived from per-token probabilities to GRPO-based RLVR and reports 2.3-4% average gains over baselines across model scales.
Discussion (0). Sign in to comment.