Prover-Verifier Games improve legibility of LLM outputs

• URL: http://arxiv.org/abs/2407.13692v2
• Date: Thu, 1 Aug 2024 17:18:54 GMT
• Title: Prover-Verifier Games improve legibility of LLM outputs
• Authors: Jan Hendrik Kirchner, Yining Chen, Harri Edwards, Jan Leike, Nat McAleese, Yuri Burda,
• Abstract summary: We study legibility in the context of solving grade-school math problems. We propose a training algorithm inspired by Prover-Verifier Game from Anil et al. We show that legibility training transfers to time-constrained humans tasked with verifying solution correctness.
• Score: 12.532113917099885
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: One way to increase confidence in the outputs of Large Language Models (LLMs) is to support them with reasoning that is clear and easy to check -- a property we call legibility. We study legibility in the context of solving grade-school math problems and show that optimizing chain-of-thought solutions only for answer correctness can make them less legible. To mitigate the loss in legibility, we propose a training algorithm inspired by Prover-Verifier Game from Anil et al. (2021). Our algorithm iteratively trains small verifiers to predict solution correctness, "helpful" provers to produce correct solutions that the verifier accepts, and "sneaky" provers to produce incorrect solutions that fool the verifier. We find that the helpful prover's accuracy and the verifier's robustness to adversarial attacks increase over the course of training. Furthermore, we show that legibility training transfers to time-constrained humans tasked with verifying solution correctness. Over course of LLM training human accuracy increases when checking the helpful prover's solutions, and decreases when checking the sneaky prover's solutions. Hence, training for checkability by small verifiers is a plausible technique for increasing output legibility. Our results suggest legibility training against small verifiers as a practical avenue for increasing legibility of large LLMs to humans, and thus could help with alignment of superhuman models.

Related papers

• Rationale-Aware Answer Verification by Pairwise Self-Evaluation [11.763229353978321]
  We show that training reliable verifiers requires ensuring the validity of rationales in addition to the correctness of the final answers. Our results suggest that training reliable verifiers requires ensuring the validity of rationales in addition to the correctness of the final answers.
  arXiv  Detail & Related papers  (2024-10-07T08:53:00Z)
• Improving LLM Reasoning through Scaling Inference Computation with Collaborative Verification [52.095460362197336]
  Large language models (LLMs) struggle with consistent and accurate reasoning. LLMs are trained primarily on correct solutions, reducing their ability to detect and learn from errors. We propose a novel collaborative method integrating Chain-of-Thought (CoT) and Program-of-Thought (PoT) solutions for verification.
  arXiv  Detail & Related papers  (2024-10-05T05:21:48Z)
• Generative Verifiers: Reward Modeling as Next-Token Prediction [29.543787728397643]
  Verifiers or reward models are often used to enhance the reasoning performance of large language models (LLMs) We propose training verifiers using the ubiquitous next-token prediction objective, jointly on verification and solution generation. We demonstrate that GenRM outperforms discriminative, DPO verifiers, and LLM-as-a-Judge.
  arXiv  Detail & Related papers  (2024-08-27T17:57:45Z)
• LLM Critics Help Catch Bugs in Mathematics: Towards a Better Mathematical Verifier with Natural Language Feedback [71.95402654982095]
  We propose Math-Minos, a natural language feedback-enhanced verifier. Our experiments reveal that a small set of natural language feedback can significantly boost the performance of the verifier.
  arXiv  Detail & Related papers  (2024-06-20T06:42:27Z)
• Small Language Models Need Strong Verifiers to Self-Correct Reasoning [69.94251699982388]
  Self-correction has emerged as a promising solution to boost the reasoning performance of large language models (LLMs) This work explores whether small (= 13B) language models (LMs) have the ability of self-correction on reasoning tasks with minimal inputs from stronger LMs.
  arXiv  Detail & Related papers  (2024-04-26T03:41:28Z)
• Learning to Check: Unleashing Potentials for Self-Correction in Large Language Models [5.463333911506443]
  We aim to enhance the self-checking capabilities of large language models (LLMs) by constructing training data for checking tasks. We propose a specialized checking format called "Step CoT Check" Experiments demonstrate that fine-tuning with the "Step CoT Check" format significantly improves the self-checking and self-correction abilities of LLMs.
  arXiv  Detail & Related papers  (2024-02-20T14:23:23Z)
• V-STaR: Training Verifiers for Self-Taught Reasoners [71.53113558733227]
  V-STaR trains a verifier using DPO that judges correctness of model-generated solutions. Running V-STaR for multiple iterations results in progressively better reasoners and verifiers.
  arXiv  Detail & Related papers  (2024-02-09T15:02:56Z)
• Understanding and Mitigating Classification Errors Through Interpretable Token Patterns [58.91023283103762]
  Characterizing errors in easily interpretable terms gives insight into whether a classifier is prone to making systematic errors. We propose to discover those patterns of tokens that distinguish correct and erroneous predictions. We show that our method, Premise, performs well in practice.
  arXiv  Detail & Related papers  (2023-11-18T00:24:26Z)
• GRACE: Discriminator-Guided Chain-of-Thought Reasoning [75.35436025709049]
  We propose Guiding chain-of-thought ReAsoning with a CorrectnEss Discriminator (GRACE) to steer the decoding process towards producing correct reasoning steps. GRACE employs a discriminator trained with a contrastive loss over correct and incorrect steps, which is used during decoding to score next-step candidates.
  arXiv  Detail & Related papers  (2023-05-24T09:16:51Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.