On scalable oversight with weak LLMs judging strong LLMs

• URL: http://arxiv.org/abs/2407.04622v2
• Date: Fri, 12 Jul 2024 16:38:12 GMT
• Title: On scalable oversight with weak LLMs judging strong LLMs
• Authors: Zachary Kenton, Noah Y. Siegel, János Kramár, Jonah Brown-Cohen, Samuel Albanie, Jannis Bulian, Rishabh Agarwal, David Lindner, Yunhao Tang, Noah D. Goodman, Rohin Shah,
• Abstract summary: We study debate, where two AI's compete to convince a judge; consultancy, where a single AI tries to convince a judge that asks questions. We use large language models (LLMs) as both AI agents and as stand-ins for human judges, taking the judge models to be weaker than agent models.
• Score: 67.8628575615614
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Scalable oversight protocols aim to enable humans to accurately supervise superhuman AI. In this paper we study debate, where two AI's compete to convince a judge; consultancy, where a single AI tries to convince a judge that asks questions; and compare to a baseline of direct question-answering, where the judge just answers outright without the AI. We use large language models (LLMs) as both AI agents and as stand-ins for human judges, taking the judge models to be weaker than agent models. We benchmark on a diverse range of asymmetries between judges and agents, extending previous work on a single extractive QA task with information asymmetry, to also include mathematics, coding, logic and multimodal reasoning asymmetries. We find that debate outperforms consultancy across all tasks when the consultant is randomly assigned to argue for the correct/incorrect answer. Comparing debate to direct question answering, the results depend on the type of task: in extractive QA tasks with information asymmetry debate outperforms direct question answering, but in other tasks without information asymmetry the results are mixed. Previous work assigned debaters/consultants an answer to argue for. When we allow them to instead choose which answer to argue for, we find judges are less frequently convinced by the wrong answer in debate than in consultancy. Further, we find that stronger debater models increase judge accuracy, though more modestly than in previous studies.

Related papers

• Raising the Stakes: Performance Pressure Improves AI-Assisted Decision Making [57.53469908423318]
  We show the effects of performance pressure on AI advice reliance when laypeople complete a common AI-assisted task. We find that when the stakes are high, people use AI advice more appropriately than when stakes are lower, regardless of the presence of an AI explanation.
  arXiv  Detail & Related papers  (2024-10-21T22:39:52Z)
• JudgeBench: A Benchmark for Evaluating LLM-based Judges [61.048125269475854]
  JudgeBench is a benchmark for evaluating LLM-based judges on challenging response pairs spanning knowledge, reasoning, math, and coding. Our comprehensive evaluation on a collection of prompted judges, fine-tuned judges, multi-agent judges, and reward models shows that JudgeBench poses a significantly greater challenge than previous benchmarks.
  arXiv  Detail & Related papers  (2024-10-16T17:58:19Z)
• Training Language Models to Win Debates with Self-Play Improves Judge Accuracy [8.13173791334223]
  We test the robustness of debate as a method of scalable oversight by training models to debate with data generated via self-play. We find that language model based evaluators answer questions more accurately when judging models optimized to win debates.
  arXiv  Detail & Related papers  (2024-09-25T05:28:33Z)
• From Calculation to Adjudication: Examining LLM judges on Mathematical Reasoning Tasks [11.01213914485374]
  We study large language models (LLMs) on mathematical reasoning tasks. Our analysis uncovers a strong correlation between judgment performance and the candidate model task performance. We show that it is possible to use statistics, such as the task performances of the individual models, to predict judgment performance.
  arXiv  Detail & Related papers  (2024-09-06T10:09:41Z)
• Judging the Judges: Evaluating Alignment and Vulnerabilities in LLMs-as-Judges [6.609843448260634]
  The LLM-as-a-judge paradigm is rapidly gaining traction as an approach to evaluating large language models. This paper focuses on a clean scenario in which inter-human agreement is high. We identify vulnerabilities in judge models, such as their sensitivity to prompt complexity and length, and a tendency toward leniency.
  arXiv  Detail & Related papers  (2024-06-18T13:49:54Z)
• Debatrix: Multi-dimensional Debate Judge with Iterative Chronological Analysis Based on LLM [51.43102092480804]
  Debatrix is an automated debate judge based on Large Language Models (LLMs) To align with real-world debate scenarios, we introduced the PanelBench benchmark, comparing our system's performance to actual debate outcomes. The findings indicate a notable enhancement over directly using LLMs for debate evaluation.
  arXiv  Detail & Related papers  (2024-03-12T18:19:47Z)
• CFMatch: Aligning Automated Answer Equivalence Evaluation with Expert Judgments For Open-Domain Question Answering [14.366087533102656]
  Question answering (QA) can only make progress if we know if an answer is correct. Current evaluation metrics to determine answer equivalence (AE) often do not align with human judgments.
  arXiv  Detail & Related papers  (2024-01-24T01:30:25Z)
• Debate Helps Supervise Unreliable Experts [33.03555781137954]
  We show that debate between two unreliable experts can help a non-expert judge more reliably identify the truth. Comparing debate to a baseline we call consultancy, where a single expert argues for only one answer which is correct half of the time, we find that debate performs significantly better. These results show that debate is a promising approach for supervising increasingly capable but potentially unreliable AI systems.
  arXiv  Detail & Related papers  (2023-11-15T05:05:40Z)
• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena [76.21004582932268]
  We examine the usage and limitations of LLM-as-a-judge, including position, verbosity, and self-enhancement biases. We then verify the agreement between LLM judges and human preferences by introducing two benchmarks: MT-bench, a multi-turn question set; and Arena, a crowdsourced battle platform.
  arXiv  Detail & Related papers  (2023-06-09T05:55:52Z)
• Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate [85.3444184685235]
  We propose a Multi-Agent Debate (MAD) framework, in which multiple agents express their arguments in the state of "tit for tat" and a judge manages the debate process to obtain a final solution. Our framework encourages divergent thinking in LLMs which would be helpful for tasks that require deep levels of contemplation.
  arXiv  Detail & Related papers  (2023-05-30T15:25:45Z)

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.