{"paper":{"title":"Prompt Interference in LLM Post-Training: Why Pass@k Optimization Can Degrade Pass@1","arxiv_id":"2602.21189","primary_cat":"cs.LG","submitted_at":"2026-02-24T18:43:08+00:00","version_compiled":"v3"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v3","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Pass@$k$ is a widely used performance metric for verifiable large language model tasks, including mathematical reasoning, code generation, and short-answer reasoning.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.04,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.09,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":1.0,"id":"s1","own":false,"role":"background","text":"It defines success if any of $k$ independently sampled solutions passes a verifier.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.05,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.41,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":1.0,"id":"s2","own":false,"role":"background","text":"This multi-sample inference metric has motivated inference-aware fine-tuning methods that directly optimize pass@$k$.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.06,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.09,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":0.99,"id":"s3","own":false,"role":"background","text":"However, prior work reports a trade-off: pass@$k$ improves while pass@1 degrades under such methods.","hedge":{"p":0.95,"label":"asserted"},"own_p":0.03,"main_claim_p":0.14,"quantitative":true,"quantitative_p":0.66,"role_probabilities":{"gap":0.01,"method":0.0,"result":0.0,"background":0.99,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":4,"p":0.84,"id":"s4","own":false,"role":"background","text":"This trade-off is practically important because pass@1 often remains a hard operational constraint due to latency and cost budgets, imperfect verifier coverage, and the need for a reliable single-shot fallback.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.08,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.17,"role_probabilities":{"gap":0.02,"method":0.0,"result":0.0,"background":0.84,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.14}},{"i":5,"p":1.0,"id":"s5","own":true,"role":"contribution","text":"We study the origin of this trade-off and provide a theoretical characterization of when pass@$k$ policy optimization can reduce pass@1 through gradient conflict induced by prompt interference.","hedge":{"p":0.91,"label":"asserted"},"own_p":0.92,"main_claim_p":0.47,"quantitative":false,"quantitative_p":0.08,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":1.0,"interpretation":0.0}},{"i":6,"p":0.74,"id":"s6","own":false,"role":"interpretation","text":"This interference arises because shared model parameters couple prompts, so improving performance on some prompts can adversely affect others.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.42,"main_claim_p":0.11,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.04,"result":0.09,"background":0.03,"limitation":0.0,"future_work":0.0,"contribution":0.1,"interpretation":0.74}},{"i":7,"p":0.5,"id":"s7","own":true,"role":"interpretation","text":"We show that pass@$k$ policy gradients can conflict with pass@1 gradients because pass@$k$ optimization implicitly reweights prompts toward low-success prompts; when these prompts are negatively interfering, their upweighting can rotate the pass@$k$ update direction away from the pass@1 direction.","hedge":{"p":0.98,"label":"asserted"},"own_p":0.89,"main_claim_p":0.7,"quantitative":false,"quantitative_p":0.24,"role_probabilities":{"gap":0.0,"method":0.02,"result":0.23,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.25,"interpretation":0.5}},{"i":8,"p":0.59,"id":"s8","own":true,"role":"method","text":"We illustrate our findings with large language model experiments on verifiable mathematical reasoning and coding tasks.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.94,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.59,"result":0.02,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.39,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":0.87,"text":"We show that pass@$k$ policy gradients can conflict with pass@1 gradients because pass@$k$ optimization implicitly reweights prompts toward low-success prompts; when these prompts are negatively interfering, their upweighting can rotate the pass@$k$ update direction away from the pass@1 direction.","location":"abstract s7","hedge_strength":{"p":0.98,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":6903,"cost_usd_reported":0.0},"compiled_at":"2026-10-14","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.05,"states_limitation":0.13,"human_participants":0.03,"controlled_comparison":0.19,"randomized_assignment":0.02,"states_quantitative_result":0.05,"compares_with_prior_results":0.76,"mentions_code_or_data_availability":0.02},"evidential_basis":{"p":0.83,"value":"observation or experiment","probabilities":{"argument":0.01,"literature":0.0,"proof_or_derivation":0.13,"analysis_of_existing_data":0.0,"observation_or_experiment":0.83,"simulation_or_computation":0.03}},"contribution_type":{"p":0.98,"value":"theoretical result","probabilities":{"review":0.0,"method_or_tool":0.0,"empirical_study":0.02,"simulation_study":0.0,"theoretical_result":0.98,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":0.98,"text":"We study the origin of this trade-off and provide a theoretical characterization of when pass@$k$ policy optimization can reduce pass@1 through gradient conflict induced by prompt interference.","location":"abstract s5"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v3"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":9,"sentences_in_abstract":9},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}