Agent-BRACE improves LLM agent performance on long-horizon partially observable tasks by 5.3-14.5% through a decoupled belief state of verbalized atomic claims with certainty labels that keeps context length constant.
arXiv preprint arXiv:2410.09724 , year=
15 Pith papers cite this work. Polarity classification is still indexing.
15
Pith papers citing it
citation-role summary
background 3
citation-polarity summary
roles
background 3representative citing papers
A verifiable empirical win rate reward combined with gradient masking enables RL training of a 7B model to reach betting-market calibration on NFL win probabilities using only outcome data.
Distinct Leaf Enumeration (DLE) replaces stochastic self-consistency sampling with deterministic traversal of a truncated decoding tree to enumerate distinct leaves, increasing coverage and reducing redundant computation while improving performance on math, coding, and reasoning benchmarks.
RLMF uses quality of model self-judgments to refine RL rankings and select training data, achieving SOTA faithful calibration while preserving accuracy and outperforming standard RL by up to 63%.
Probability calibration applied to LLM evaluator judgments reduces preference coupling gamma by 20-49% and Jensen-Shannon divergence by 45-67% in a within-subjects experiment with N=5.
CHAL is a multi-agent dialectic system that performs structured belief optimization over defeasible domains using Bayesian-inspired graph representations and configurable meta-cognitive value system hyperparameters.
Unsupervised single-generation confidence calibration for reasoning LLMs via offline self-consistency proxy distillation outperforms baselines on math and QA tasks and improves selective prediction.
GroupDPO decouples group-wise preference optimization during backpropagation to cut peak memory while keeping the same gradients, allowing larger groups and consistent gains over single-pair DPO plus an NLL term on positives.
CURA improves calibration of clinical LM risk predictions by combining individual error alignment with neighborhood-based soft labels without harming discrimination on MIMIC-IV tasks.
DCPO decouples reasoning and calibration in RLVR, matching GRPO accuracy while reducing overconfidence via a claimed fundamental gradient conflict.
C3RL is a new RL algorithm combining correctness, calibration, and reference accuracy rewards to improve LLM confidence calibration, enabling CAS to outperform majority voting with up to 12.33x lower inference cost.
Supervised fine-tuning degrades the correlation between confidence scores and output quality in language models, driven by factors like training distribution similarity rather than true quality.
Budget-matched evaluations and partial-prompt contamination probes show several RLVR reasoning gaps shrink or vanish once budgets, prompts, and dataset versions are controlled.
Headache specialists preferred their own literature summaries over those from Sonnet, GPT-4o, and Llama 3.1 in a blinded evaluation, though AI summaries were sometimes indistinguishable.
Proposes a five-bucket taxonomy of LLM harms and calls for dynamic auditing, but the systematic review behind it is not reproducible and contains mismatched citations.
citing papers explorer
-
Agent-BRACE: Decoupling Beliefs from Actions in Long-Horizon Tasks via Verbalized State Uncertainty
Agent-BRACE improves LLM agent performance on long-horizon partially observable tasks by 5.3-14.5% through a decoupled belief state of verbalized atomic claims with certainty labels that keeps context length constant.
-
Verifiable Rewards for Calibrated Probabilistic Forecasting
A verifiable empirical win rate reward combined with gradient masking enables RL training of a 7B model to reach betting-market calibration on NFL win probabilities using only outcome data.
-
Efficient Test-Time Inference via Deterministic Exploration of Truncated Decoding Trees
Distinct Leaf Enumeration (DLE) replaces stochastic self-consistency sampling with deterministic traversal of a truncated decoding tree to enumerate distinct leaves, increasing coverage and reducing redundant computation while improving performance on math, coding, and reasoning benchmarks.
-
Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs
RLMF uses quality of model self-judgments to refine RL rankings and select training data, achieving SOTA faithful calibration while preserving accuracy and outperforming standard RL by up to 63%.
-
Calibrating the Evaluator: Does Probability Calibration Mitigate Preference Coupling in LLM Agent Feedback Loops?
Probability calibration applied to LLM evaluator judgments reduces preference coupling gamma by 20-49% and Jensen-Shannon divergence by 45-67% in a within-subjects experiment with N=5.
-
CHAL: Council of Hierarchical Agentic Language
CHAL is a multi-agent dialectic system that performs structured belief optimization over defeasible domains using Bayesian-inspired graph representations and configurable meta-cognitive value system hyperparameters.
-
Unsupervised Confidence Calibration for Reasoning LLMs from a Single Generation
Unsupervised single-generation confidence calibration for reasoning LLMs via offline self-consistency proxy distillation outperforms baselines on math and QA tasks and improves selective prediction.
-
GroupDPO: Memory efficient Group-wise Direct Preference Optimization
GroupDPO decouples group-wise preference optimization during backpropagation to cut peak memory while keeping the same gradients, allowing larger groups and consistent gains over single-pair DPO plus an NLL term on positives.
-
CURA: Clinical Uncertainty Risk Alignment for Language Model-Based Risk Prediction
CURA improves calibration of clinical LM risk predictions by combining individual error alignment with neighborhood-based soft labels without harming discrimination on MIMIC-IV tasks.
-
Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards
DCPO decouples reasoning and calibration in RLVR, matching GRPO accuracy while reducing overconfidence via a claimed fundamental gradient conflict.
-
Scaling with Confidence: Calibrating Confidence of LLMs for Adaptive Test Time Scaling
C3RL is a new RL algorithm combining correctness, calibration, and reference accuracy rewards to improve LLM confidence calibration, enabling CAS to outperform majority voting with up to 12.33x lower inference cost.
-
Confident in a Confidence Score: Investigating the Sensitivity of Confidence Scores to Supervised Fine-Tuning
Supervised fine-tuning degrades the correlation between confidence scores and output quality in language models, driven by factors like training distribution similarity rather than true quality.
-
Position: The Hidden Costs and Measurement Gaps of Reinforcement Learning with Verifiable Rewards
Budget-matched evaluations and partial-prompt contamination probes show several RLVR reasoning gaps shrink or vanish once budgets, prompts, and dataset versions are controlled.
-
Ten Headache Specialists versus Artificial Intelligence for Clinical Literature Summarization: A Critical Evaluation and Comparison
Headache specialists preferred their own literature summaries over those from Sonnet, GPT-4o, and Llama 3.1 in a blinded evaluation, though AI summaries were sometimes indistinguishable.
-
LLM Harms: A Taxonomy and Discussion
Proposes a five-bucket taxonomy of LLM harms and calls for dynamic auditing, but the systematic review behind it is not reproducible and contains mismatched citations.