MétaCan
Menu
Back to cohort

Action Elimination through Hallucination Detection: A Reinforcement Learning Approach for LLM Fine-Tuning

2025· article· W4417003494 on OpenAlex
Jeong Ho Park, Song Jae Lee, Kyutae Cho

Why this work is in the frame

A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.

affAt least one author lists a Canadian institution in the pinned OpenAlex snapshot.

Bibliographic record

Venuenot available
Typearticle
Language
FieldComputer Science
TopicAdversarial Robustness in Machine Learning
Canadian institutionsNexen (Canada)
Fundersnot available
KeywordsReinforcement learningAction (physics)Focus (optics)Filter (signal processing)Noise (video)Reinforcement

Abstract

fetched live from OpenAlex

The present paper proposes a novel fine-tuning framework for large language models (LLMs) in unlabeled data settings by integrating reinforcement learning (RL) with an action elimination mechanism. The proposed method is designed to detect and filter out responses that are prone to hallucinations prior to policy updates, thereby enabling RL to focus exclusively on high-quality outputs. This has been demonstrated to reduce noise during training, improve reward signal quality, and accelerate convergence. Experiments conducted on mathematics and reasoning benchmarks demonstrate consistent performance improvements across multiple model architectures. In particular, the application of action elimination has been demonstrated to engender supplementary gains in both domain-specific and cross-domain reasoning tasks. The findings of this study suggest that substantial performance enhancements can be attained in the absence of labelled data.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.002
metaresearch head score (Gemma)0.001
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesMeta-epidemiology (narrow), Science and technology studies
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: Simulation or modeling
GenreCandidate signal: Methods · Consensus signal: none
Teacher disagreement score0.736
Threshold uncertainty score1.000

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0020.001
Meta-epidemiology (narrow)0.0010.001
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0010.002
Science and technology studies0.0020.000
Scholarly communication0.0010.003
Open science0.0010.001
Research integrity0.0000.001
Insufficient payload (model declined to judge)0.0000.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.037
GPT teacher head0.317
Teacher spread0.280 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it