{"paper":{"title":"MemBuilder: Reinforcing LLMs for Long-Term Memory Construction via Attributed Dense Rewards","license":"http://creativecommons.org/licenses/by/4.0/","headline":"MemBuilder trains a 4B model with dense rewards to outperform closed-source LLMs on long-term dialogue memory.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Fuming Lai, Shaobing Lian, Yanghui Rao, Zhiyu Shen, Ziming Wu","submitted_at":"2026-01-09T02:44:37Z","abstract_excerpt":"Maintaining consistency in long-term dialogues remains a fundamental challenge for LLMs, as standard retrieval mechanisms often fail to capture the temporal evolution of historical states. While memory-augmented frameworks offer a structured alternative, current systems rely on static prompting of closed-source models or suffer from ineffective training paradigms with sparse rewards. We introduce MemBuilder, a reinforcement learning framework that trains models to orchestrate multi-dimensional memory construction with attributed dense rewards. MemBuilder addresses two key challenges: (1) Spars"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experimental results show that MemBuilder enables a 4B-parameter model to outperform state-of-the-art closed-source baselines, exhibiting strong generalization across long-term dialogue benchmarks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That synthetic session-level question generation produces unbiased, representative dense rewards that accurately measure and improve real-world memory construction quality without introducing artifacts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MemBuilder trains 4B-parameter models with attributed dense rewards to outperform closed-source baselines on long-term dialogue memory tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MemBuilder trains a 4B model with dense rewards to outperform closed-source LLMs on long-term dialogue memory.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ccfbef3e30dbb23547d333e01a49c691f8be3498c592a377ffa851f87fa0fcca"},"source":{"id":"2601.05488","kind":"arxiv","version":4},"verdict":{"id":"0838669c-3178-4531-b8bf-c0848244cc2f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T16:53:01.591957Z","strongest_claim":"Experimental results show that MemBuilder enables a 4B-parameter model to outperform state-of-the-art closed-source baselines, exhibiting strong generalization across long-term dialogue benchmarks.","one_line_summary":"MemBuilder trains 4B-parameter models with attributed dense rewards to outperform closed-source baselines on long-term dialogue memory tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That synthetic session-level question generation produces unbiased, representative dense rewards that accurately measure and improve real-world memory construction quality without introducing artifacts.","pith_extraction_headline":"MemBuilder trains a 4B model with dense rewards to outperform closed-source LLMs on long-term dialogue memory."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2601.05488/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"05d0f612b60166ebd6c38efa8142ad949ea10fca898c3dcf954e947f08f7fc1b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}