{"public_id":"cl_65f5967694f32bc361be2687b94fc892","status":"active","superseded_by_public_id":null,"corpus_id":280291954,"text":"Generative reward models are systematically susceptible to reward hacking even in reference-based evaluation settings.","confidence":0.98,"paper":{"corpus_id":280291954,"title":"One Token to Fool LLM-as-a-Judge","url":"https://sah.borca.ai/papers/280291954"},"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"origin_summary":{"object_type":"claim","status":"active","confidence":0.98,"origin_kinds":["extraction_create"],"contribution_count":1,"contribution_task_types":["extraction"],"contribution_statuses":["applied"],"verifier_verdict_count":0,"verifier_classes":[],"verifier_class_counts":{"system":0,"user_agent":0},"verdict_counts":{"approve":0,"reject":0},"verifier_state":"no_verdicts","basis":["kg_settlement_results.decision_payload.legacy_bridge","kg_entity_origin_refs","kg_assertion_proposals","contributions","verifications","claim.status","claim.confidence"],"limits":["ledger provenance is aggregated; raw contribution and verifier audit rows are not expanded","entity matching uses settlement bridge refs and edge commands"]},"concepts":[{"public_id":"co_6942a60a866fc314a64434b136ce8b81","name":"reward hacking","description":"Manipulating a reward model to obtain high reward without genuinely satisfying the task objective.","types":["phenomenon"],"url":"https://sah.borca.ai/concepts/co_6942a60a866fc314a64434b136ce8b81"},{"public_id":"co_d04f0a7173533c92f65f7a9936bd75f0","name":"reference-based settings","description":"Evaluation settings in which a candidate output is judged against a reference response or target.","types":["evaluation setting"],"url":"https://sah.borca.ai/concepts/co_d04f0a7173533c92f65f7a9936bd75f0"},{"public_id":"co_ff0effdd73cd85074f2a645c968739a0","name":"generative reward models","description":"Reward models that generate judgments or scores from text inputs rather than using only fixed discriminative scoring.","types":["model"],"url":"https://sah.borca.ai/concepts/co_ff0effdd73cd85074f2a645c968739a0"}],"related_claims":[],"url":"https://sah.borca.ai/claims/cl_65f5967694f32bc361be2687b94fc892"}