Research map / Large language models / Language, safety and interpretability
Latent Debate: A Surrogate Framework for Interpreting LLM Thinking towards Binary Decisions
Lihu Chen, Xiang Yin, Francesca Toni
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces latent debate, a structured surrogate framework that interprets LLM True/False predictions via hidden supporting and attacking latent arguments and their interactions.
Problem
Understanding the internal thinking process of LLMs and the causes of hallucinations on binary True/False prediction tasks remains a key challenge.
Approach
Treats an LLM's internal representations as a debate among latent arguments (supporting and attacking) and uses a structured surrogate model to interpret how these interactions produce the binary output, unlike human debates that are explicit.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add token-wise semantic weights to initial strengths (ablation) | Latent Debate · Prediction consistency, Llama-2-13B | consistency (%): 95.0 -> 96.6 | Improved |
| Add token-wise semantic weights (ablation) | Latent Debate · Prediction consistency, Llama-3.1-8B | consistency (%): 80.1 -> 83.6 | Improved |
| Use quadratic connections instead of simple topology (ablation) | Latent Debate · Prediction consistency, Llama-2-13B | consistency (%): 96.6 -> 79.6 | Hurt |
| Use richer graph topologies (adjacent grid, quadratic, fully quadratic) (ablation) | Latent Debate · Consistency, Llama-3.1-8B avg of 6 | consistency (%): 84.27 -> 78.50 / 58.60 / 47.80 | Hurt |
| Replace Logit Lens with Tuned Lens as argument interpreter (ablation) | Latent Debate · Consistency, 3 datasets | consistency (%): 86.60 -> 93.20 | Improved |
| Replace Logit Lens with Tuned Lens as argument interpreter (ablation) | Latent Debate Detector · Hallucination detection, 3 datasets | AUROC: 86.54 -> 67.03 | Hurt |
| Use latent debate aggregation over simple average/majority vote | Latent Debate · Prediction consistency, Llama-2-13B | consistency (%): 90.0 -> 96.6 | Improved |
| Use latent debate with Mistral-7B vs majority voting | Latent Debate · Prediction consistency, Mistral-7B | consistency (%): 92.3 -> 92.4 | No clear effect |
| Use rightmost top argument instead of full debate (ablation) | Single-argument baseline · Consistency, Llama-3.1-8B | consistency (%): 83.57 -> 87.03 | Improved |
| Use debate features instead of single top-right argument for hallucination detection | Latent Debate Detector · Hallucination detection, Llama-3.1-8B | AUROC: 0.58 -> 0.79 | Improved |
| Remove lower layers (first 3 to 12) from latent debate (ablation) | Latent Debate Detector · Hallucination detection | AUROC: CounterFact 72.44 -> down 2.55 (first 3); TruthfulQA down 2.07 (first 12) | Mixed |
| Use latent debate features instead of SAPLMA across tasks | Latent Debate Detector · Cross-task detection (trained on Common Claim) | AUROC: 68.89 -> 75.26 | Improved |
Closest papers by meaning
- Auditing Meta-Cognitive Hallucinations in Reasoning Large Language Models (NeurIPS 2025 · similarity 0.90)
- LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations (ICLR 2025 · similarity 0.90)
- Joint Evaluation of Answer and Reasoning Consistency for Hallucination Detection in Large Reasoning Models (AAAI 2026 · similarity 0.90)
- Mitigating Hallucinations in Large Language Models via Causal Reasoning (AAAI 2026 · similarity 0.90)
- In-Context Sharpness as Alerts: An Inner Representation Perspective for Hallucination Mitigation (ICML 2024 · similarity 0.89)
- Why and How LLMs Hallucinate: Connecting the Dots with Subsequence Associations (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
