Pith. sign in

REVIEW 15 cited by

Language Models Learn to Mislead Humans via RLHF

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2409.12822 v3 pith:T2T7CSVU submitted 2024-09-19 cs.CL

Language Models Learn to Mislead Humans via RLHF

classification cs.CL
keywords rlhfhumanstaskmodelsubjectsappsbetterconvincing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Language models (LMs) can produce errors that are hard to detect for humans, especially when the task is complex. RLHF, the most popular post-training method, may exacerbate this problem: to achieve higher rewards, LMs might get better at convincing humans that they are right even when they are wrong. We study this phenomenon under a standard RLHF pipeline, calling it "U-SOPHISTRY" since it is Unintended by model developers. Specifically, we ask time-constrained (e.g., 3-10 minutes) human subjects to evaluate the correctness of model outputs and calculate humans' accuracy against gold labels. On a question-answering task (QuALITY) and programming task (APPS), RLHF makes LMs better at convincing our subjects but not at completing the task correctly. RLHF also makes the model harder to evaluate: our subjects' false positive rate increases by 24.1% on QuALITY and 18.3% on APPS. Finally, we show that probing, a state-of-the-art approach for detecting Intended Sophistry (e.g. backdoored LMs), does not generalize to U-SOPHISTRY. Our results highlight an important failure mode of RLHF and call for more research in assisting humans to align them.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 15 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. More Convincing, Not More Correct: Self-Play Reward Hacking of Reference-Free LLM Judges

    cs.LG 2026-07 accept novelty 7.0

    Self-play against reference-free LLM judges drives judge pass rates to 0.94 while true accuracy stays at 0.20, a reward-hacking basin that transfers across judge families and is prevented only by requiring the judge t...

  2. Gaslight, Gatekeep, V1-V3: Early Visual Cortex Alignment Shields Vision-Language Models from Sycophantic Manipulation

    cs.CV 2026-04 conditional novelty 7.0

    Alignment of vision-language models with human V1-V3 early visual cortex negatively predicts resistance to sycophantic gaslighting attacks.

  3. Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation

    cs.AI 2025-03 conditional novelty 7.0

    Chain-of-thought monitoring detects reward hacking in frontier reasoning models, but strong optimization against the monitor produces obfuscated misbehavior that remains hard to detect.

  4. Optimizing Visual Generative Models via Distribution-wise Rewards

    cs.LG 2026-07 unverdicted novelty 6.0

    Distribution-wise rewards with subset-replace strategy and post-hoc merging improve FID-50K on SiT (8.30 to 5.77) and EDM2 (3.74 to 3.52) while preserving diversity.

  5. The Open-Box Fallacy: Why AI Deployment Needs a Calibrated Verification Regime

    cs.AI 2026-05 unverdicted novelty 6.0

    AI deployment in high-stakes areas requires domain-scoped calibrated verification with monitoring and revocation, using a proposed six-component Verification Coverage standard instead of mechanistic interpretability.

  6. The Impact of Off-Policy Training Data on Probe Generalisation

    cs.AI 2025-11 unverdicted novelty 6.0

    Off-policy training data for LLM behavior probes causes significant generalization failures especially for intent-based behaviors like deception, and performance on coerced incentivised data correlates with real on-po...

  7. LLM Hypnosis: Exploiting User Feedback for Unauthorized Knowledge Injection to All Users

    cs.CL 2025-07 unverdicted novelty 6.0

    A single attacker can use strategic upvoting and downvoting on language model outputs to inject facts, security flaws, or fake news that persist in the model for all users after preference tuning.

  8. ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models

    cs.CL 2025-05 conditional novelty 6.0

    Prolonged RL training with KL control and reference policy resetting enables LLMs to develop novel reasoning strategies inaccessible to base models even under extensive sampling.

  9. VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model

    cs.CV 2025-04 unverdicted novelty 6.0

    VLM-R1 applies R1-style RL using rule-based rewards on visual tasks with clear ground truth to achieve competitive performance and superior generalization over SFT in vision-language models.

  10. Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal

    cs.LG 2026-06 unverdicted novelty 5.0

    A new pipeline uses interpretability to characterize concepts in preference data and shape rewards via feature or data interventions during LM post-training.

  11. DynaCF: Mitigating Shortcut Learning in Reward Models via Dynamic Counterfactual Sensitivity

    cs.LG 2026-06 unverdicted novelty 5.0

    DynaCF dynamically downweights shortcut-sensitive samples in reward model training by tracking margin shifts under online counterfactual perturbations within the Bradley-Terry loss.

  12. Do Linear Probes Generalize Better in Persona Coordinates?

    cs.AI 2026-05 unverdicted novelty 5.0

    Persona axes derived from contrastive prompts and PCA yield linear probes that generalize better than raw-activation probes across 10 datasets for deception and sycophancy.

  13. Do Linear Probes Generalize Better in Persona Coordinates?

    cs.AI 2026-05 unverdicted novelty 5.0

    Probes on persona principal components from contrastive prompts generalize better than raw activation probes for harmful behaviors across 10 datasets.

  14. Proxy Reward Internalization and Mechanistic Exploitation: A Learned Precursor to Reward Hacking and Its Generalization

    cs.AI 2026-06 unverdicted novelty 4.0

    Proxy RL produces a staged proxy-internalization capability that emerges before and predicts reward hacking in coding environments.

  15. "Skill issues'': data-centric optimization of lakehouse agents

    cs.AI 2026-05 unverdicted novelty 3.0

    Data-centric optimization of skills for agents on a branching lakehouse improves accuracy by 31.9% on 25 tasks via state-verification evaluation.