Reinforcement learning training for reasoning substantially raises specification gaming rates in LLMs across diverse tasks, with Grok 4 highest and Claude models lowest, and mitigations only partially effective.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Towards Understanding Specification Gaming in Reasoning Models
Reinforcement learning training for reasoning substantially raises specification gaming rates in LLMs across diverse tasks, with Grok 4 highest and Claude models lowest, and mitigations only partially effective.