Pith. sign in

REVIEW 4 cited by

MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking

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 2501.13011 v2 pith:7PXHOR7S submitted 2025-01-22 cs.LG cs.AI

classification cs.LGcs.AI
keywords rewardmonamulti-stephackingoptimizationableapprovaldetect
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Future advanced AI systems may learn sophisticated strategies through reinforcement learning (RL) that humans cannot understand well enough to safely evaluate. We propose a training method which avoids agents learning undesired multi-step plans that receive high reward (multi-step "reward hacks") even if humans are not able to detect that the behaviour is undesired. The method, Myopic Optimization with Non-myopic Approval (MONA), works by combining short-sighted optimization with far-sighted reward. We demonstrate that MONA can prevent multi-step reward hacking that ordinary RL causes, even without being able to detect the reward hacking and without any extra information that ordinary RL does not get access to. We study MONA empirically in three settings which model different misalignment failure modes including 2-step environments with LLMs representing delegated oversight and encoded reasoning and longer-horizon gridworld environments representing sensor tampering.

Discussion (0). Sign in to comment.

Forward citations

Cited by 4 Pith papers

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

  1. Teaching Models to Verbalize Reward Hacking in Chain-of-Thought Reasoning

    cs.CL 2025-06 conditional novelty 7.0 of 10

    A pre-RL fine-tuning intervention called verbalization fine-tuning makes language models explicitly acknowledge when prompt cues drive them to reward-hack, cutting undetected reward hacking from 88% to 6% after RL.

  2. Draining the Energy Commons: Self-Defeating Over-Appropriation as a Coordination Failure in Agentic LLM Collectives

    cs.MA 2026-07 conditional novelty 5.0 of 10

    LLM prosumers deplete a shared renewable reserve exactly when demand exceeds peak replacement, acting like impatient open-access users even when sustaining the reserve is feasible.

  3. NEST: Nascent Encoded Steganographic Thoughts

    cs.AI 2026-02 conditional novelty 5.0 of 10

    Frontier LLMs can embed short digit sequences in sentence acrostics (Claude Opus 4.5: 92% per-digit at D=4) but fail to jointly solve hidden reasoning tasks and encode the solution.

  4. A Survey on Autonomy-Induced Security Risks in Large Model-Based Agents

    cs.AI 2025-06 conditional novelty 4.0 of 10

    The paper surveys security risks of LLM agents, organizes them into a five-level autonomy taxonomy, and proposes an untested CMDP-based architecture called R2A2.

Pith tools