Reverse Engineering Human Preferences with Reinforcement Learning

• URL: http://arxiv.org/abs/2505.15795v1
• Date: Wed, 21 May 2025 17:48:16 GMT
• Title: Reverse Engineering Human Preferences with Reinforcement Learning
• Authors: Lisa Alazraki, Tan Yi-Chern, Jon Ander Campos, Maximilian Mozes, Marek Rei, Max Bartolo,
• Abstract summary: Large Language Models (LLMs) are routinely evaluated by other LLMs trained to predict human preferences.<n>Previous work shows that the answers generated by a candidate-LLM can be edited post hoc to maximise the score assigned to them by a judge-LLM.<n>We adopt a different approach and use the signal provided by judge-LLMs as a reward to adversarially tune models.
• Score: 14.508050809497847
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: The capabilities of Large Language Models (LLMs) are routinely evaluated by other LLMs trained to predict human preferences. This framework--known as LLM-as-a-judge--is highly scalable and relatively low cost. However, it is also vulnerable to malicious exploitation, as LLM responses can be tuned to overfit the preferences of the judge. Previous work shows that the answers generated by a candidate-LLM can be edited post hoc to maximise the score assigned to them by a judge-LLM. In this study, we adopt a different approach and use the signal provided by judge-LLMs as a reward to adversarially tune models that generate text preambles designed to boost downstream performance. We find that frozen LLMs pipelined with these models attain higher LLM-evaluation scores than existing frameworks. Crucially, unlike other frameworks which intervene directly on the model's response, our method is virtually undetectable. We also demonstrate that the effectiveness of the tuned preamble generator transfers when the candidate-LLM and the judge-LLM are replaced with models that are not used during training. These findings raise important questions about the design of more reliable LLM-as-a-judge evaluation settings. They also demonstrate that human preferences can be reverse engineered effectively, by pipelining LLMs to optimise upstream preambles via reinforcement learning--an approach that could find future applications in diverse tasks and domains beyond adversarial attacks.

Related papers

• Bayesian Teaching Enables Probabilistic Reasoning in Large Language Models [50.16340812031201]
  We show that large language models (LLMs) do not update their beliefs as expected from the Bayesian framework.<n>We teach the LLMs to reason in a Bayesian manner by training them to mimic the predictions of an optimal Bayesian model.
  arXiv  Detail & Related papers  (2025-03-21T20:13:04Z)
• LLM-Powered Preference Elicitation in Combinatorial Assignment [17.367432304040662]
  We study the potential of large language models (LLMs) as proxies for humans to simplify preference elicitation (PE) in assignment.<n>We propose a framework for LLM proxies that can work in tandem with SOTA ML-powered preference elicitation schemes.<n>We experimentally evaluate the efficiency of LLM proxies against human queries in the well-studied course allocation domain.
  arXiv  Detail & Related papers  (2025-02-14T17:12:20Z)
• Fairer Preferences Elicit Improved Human-Aligned Large Language Model Judgments [41.25558612970942]
  We show that large language models (LLMs) exhibit preference biases and worrying sensitivity to prompt designs. Motivated by this phenomenon, we propose an automatic Zero-shot Evaluation-oriented Prompt Optimization framework, ZEPO.
  arXiv  Detail & Related papers  (2024-06-17T09:48:53Z)
• Is LLM-as-a-Judge Robust? Investigating Universal Adversarial Attacks on Zero-shot LLM Assessment [8.948475969696075]
  Large Language Models (LLMs) are powerful zero-shot assessors used in real-world situations such as assessing written exams and benchmarking systems. We show that short universal adversarial phrases can be deceived to judge LLMs to predict inflated scores. It is found that judge-LLMs are significantly more susceptible to these adversarial attacks when used for absolute scoring.
  arXiv  Detail & Related papers  (2024-02-21T18:55:20Z)
• Are Large Language Models Good Prompt Optimizers? [65.48910201816223]
  We conduct a study to uncover the actual mechanism of LLM-based Prompt Optimization. Our findings reveal that the LLMs struggle to identify the true causes of errors during reflection, tending to be biased by their own prior knowledge. We introduce a new "Automatic Behavior Optimization" paradigm, which directly optimize the target model's behavior in a more controllable manner.
  arXiv  Detail & Related papers  (2024-02-03T09:48:54Z)
• PiCO: Peer Review in LLMs based on the Consistency Optimization [48.48819141999387]
  We use peer-review mechanisms to measure large language models (LLMs) automatically.<n>We formalize it as a constrained optimization problem, intending to maximize the consistency of each LLM's capabilities and scores.<n>We propose three metrics called PEN, CIN, and LIS to evaluate the gap in aligning human rankings.
  arXiv  Detail & Related papers  (2024-02-02T18:49:26Z)
• Evaluating Large Language Models at Evaluating Instruction Following [54.49567482594617]
  We introduce a challenging meta-evaluation benchmark, LLMBar, designed to test the ability of an LLM evaluator in discerning instruction-following outputs. We discover that different evaluators exhibit distinct performance on LLMBar and even the highest-scoring ones have substantial room for improvement.
  arXiv  Detail & Related papers  (2023-10-11T16:38:11Z)
• LLMRec: Benchmarking Large Language Models on Recommendation Task [54.48899723591296]
  The application of Large Language Models (LLMs) in the recommendation domain has not been thoroughly investigated. We benchmark several popular off-the-shelf LLMs on five recommendation tasks, including rating prediction, sequential recommendation, direct recommendation, explanation generation, and review summarization. The benchmark results indicate that LLMs displayed only moderate proficiency in accuracy-based tasks such as sequential and direct recommendation.
  arXiv  Detail & Related papers  (2023-08-23T16:32:54Z)
• On Learning to Summarize with Large Language Models as References [101.79795027550959]
  Large language models (LLMs) are favored by human annotators over the original reference summaries in commonly used summarization datasets. We study an LLM-as-reference learning setting for smaller text summarization models to investigate whether their performance can be substantially improved.
  arXiv  Detail & Related papers  (2023-05-23T16:56:04Z)

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.