Previous work assigned debaters/consultants an answer to argue for.

This dataset includes over 3.5 million documents with rich metadata, making it one of the most extensive collections of debate evidence.

The exponential growth of digital content presents significant challenges for content safety. Current moderation systems, often based on single models or fixed pipelines, exhibit limitations in identifying implicit risks and providing interpretable judgment processes. To address these issues, we propose Aetheria, a multimodal interpretable content safety framework based on multi-agent debate and collaboration.Employing a collaborative architecture of five core agents, Aetheria conducts in-depth analysis and adjudication of multimodal content through a dynamic, mutually persuasive debate mechanism, which is grounded by RAG-based knowledge retrieval.Comprehensive experiments on our proposed benchmark (AIR-Bench) validate that Aetheria not only generates detailed and traceable audit reports but also demonstrates significant advantages over baselines in overall content safety accuracy, especially in the identification of implicit risks. This framework establishes a transparent and interpretable paradigm, significantly advancing the field of trustworthy AI content moderation.

The core premise of AI debate as a scalable oversight technique is that it is harder to lie convincingly than to refute a lie, enabling the judge to identify the correct position. Yet, existing debate experiments have relied on datasets with ground truth, where lying is reduced to defending an incorrect proposition. This overlooks a subjective dimension: lying also requires the belief that the claim defended is false. In this work, we apply debate to subjective questions and explicitly measure large language models' prior beliefs before experiments. Debaters were asked to select their preferred position, then presented with a judge persona deliberately designed to conflict with their identified priors. This setup tested whether models would adopt sycophantic strategies, aligning with the judge's presumed perspective to maximize persuasiveness, or remain faithful to their prior beliefs. We implemented and compared two debate protocols, sequential and simultaneous, to evaluate potential systematic biases. Finally, we assessed whether models were more persuasive and produced higher-quality arguments when defending positions consistent with their prior beliefs versus when arguing against them. Our main findings show that models tend to prefer defending stances aligned with the judge persona rather than their prior beliefs, sequential debate introduces significant bias favoring the second debater, models are more persuasive when defending positions aligned with their prior beliefs, and paradoxically, arguments misaligned with prior beliefs are rated as higher quality in pairwise comparison. These results can inform human judges to provide higher-quality training signals and contribute to more aligned AI systems, while revealing important aspects of human-AI interaction regarding persuasion dynamics in language models.

Accurate detection of errors in large language models (LLM) responses is central to the success of scalable oversight, or providing effective supervision to superhuman intelligence. Yet, self-diagnosis is often unreliable on complex tasks unless aided by reliable external feedback. Multi-agent debate (MAD) seems to be a natural alternative to external feedback: multiple LLMs provide complementary perspectives and cross-checks for error detection. However, prior MAD protocols frame debate as a zero-sum game, where the debaters compete to win the game instead of seeking the truth. Consequently, it leads to debate hacking: debaters tend to mislead the judge by misinterpreting the task or presenting overconfident claims, which introduce more mistakes and underperform single-agent methods. To mitigate the issue, we introduce a new collaborative MAD protocol, termed ColMAD, that reframes MAD as a non-zero sum game. Specifically, ColMAD encourages multiple agents to criticize each other in a supportive way, such that they can complement the missing points of each other. Therefore, the judge agent can make a more informative conclusion based on more comprehensive evidence. Empirically, we show that ColMAD significantly outperforms previous competitive MAD by 19% and brings non-trivial improvements over single-agent methods in error detection.

Large Language Models (LLMs) optimized to output truthful answers often overfit, producing brittle reasoning that fails to generalize. While persuasion-based optimization has shown promise in debate settings, it has not been systematically compared against mainstream truth-based approaches. We introduce DebateQD, a minimal Quality-Diversity (QD) evolutionary algorithm that evolves diverse debate strategies across different categories (rationality, authority, emotional appeal, etc.) through tournament-style competitions where two LLMs debate while a third judges. Unlike previously proposed methods that require a population of LLMs, our approach maintains diversity of opponents through prompt-based strategies within a single LLM architecture, making it more accessible for experiments while preserving the key benefits of population-based optimization. In contrast to prior work, we explicitly isolate the role of the optimization objective by fixing the debate protocol and swapping only the fitness function: persuasion rewards strategies that convince the judge irrespective of truth, whereas truth rewards collaborative correctness. Across three model scales (7B, 32B, 72B parameters) and multiple dataset sizes from the QuALITY benchmark, persuasion-optimized strategies achieve up to 13.94% smaller train-test generalization gaps, while matching or exceeding truth optimization's test performance. These results provide the first controlled evidence that competitive pressure to persuade, rather than seek the truth collaboratively, fosters more transferable reasoning skills, offering a promising path for improving LLM generalization.

Please refer to table 1. The harmful keywords are with red and the unharmful keywords are withgreen background color.

This dataset includes over 3.5 million documents with rich metadata, making it one of the most extensive collections of debate evidence.