{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:3AKEZE3TAPIIHZYC5RBY7ZDHZN","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"909f67bdf57361e1e01dbde35b145d65f5a3e5e6d9083acc5b7b887a65a2b28d","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-03T03:11:50Z","title_canon_sha256":"1c5f2eb4f7d03e2225f06250d7bc510eab14efdecea0d98d161a35291af95019"},"schema_version":"1.0","source":{"id":"2507.02259","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2507.02259","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"arxiv_version","alias_value":"2507.02259v1","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.02259","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"pith_short_12","alias_value":"3AKEZE3TAPII","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"3AKEZE3TAPIIHZYC","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"3AKEZE3T","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:a4bb4eb0b70b7a614dd1b81e215a99b5597d42617e9b4d41be2caa3057e83e7b","target":"graph","created_at":"2026-05-17T23:38:52Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"MemAgent has demonstrated superb long-context capabilities, being able to extrapolate from an 8K context trained on 32K text to a 3.5M QA task with performance loss < 5% and achieves 95%+ in 512K RULER test."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the overwrite memory strategy combined with multi-conversation RL training will continue to prevent performance degradation when scaling far beyond the 32K training length without additional mechanisms or data."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"MemAgent uses multi-conversation RL to train a memory agent that reads text in segments and overwrites memory, extrapolating from 8K training to 3.5M token QA with under 5% loss and 95%+ on 512K RULER."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"MemAgent lets LLMs handle millions of tokens by segmenting input and overwriting memory after RL training on 32K texts."}],"snapshot_sha256":"37d6df6170b4e6d89a70129ae69baaeb847622fd5e1472e0ab5d8c9f2783130d"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"57dc516578b067252ffe60da63e7f3913ced605f33dcf1ea6b02242da768c781"},"paper":{"abstract_excerpt":"Despite improvements by length extrapolation, efficient attention and memory modules, handling infinitely long documents with linear complexity without performance degradation during extrapolation remains the ultimate challenge in long-text processing. We directly optimize for long-text tasks in an end-to-end fashion and introduce a novel agent workflow, MemAgent, which reads text in segments and updates the memory using an overwrite strategy. We extend the DAPO algorithm to facilitate training via independent-context multi-conversation generation. MemAgent has demonstrated superb long-context","authors_text":"Hao Zhou, Hongli Yu, Jiangjie Chen, Jiangtao Feng, Jingjing Liu, Mingxuan Wang, Qiying Yu, Tinghong Chen, Weinan Dai, Wei-Ying Ma, Ya-Qin Zhang","cross_cats":["cs.AI","cs.LG"],"headline":"MemAgent lets LLMs handle millions of tokens by segmenting input and overwriting memory after RL training on 32K texts.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-03T03:11:50Z","title":"MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent"},"references":{"count":65,"internal_anchors":33,"resolved_work":65,"sample":[{"cited_arxiv_id":"2404.06654","doi":"","is_internal_anchor":true,"ref_index":1,"title":"RULER: What's the Real Context Size of Your Long-Context Language Models?","work_id":"c0bc4689-3ce8-4e3d-9442-bd74869445bb","year":2024},{"cited_arxiv_id":"1809.09600","doi":"","is_internal_anchor":true,"ref_index":2,"title":"HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering","work_id":"c87d7e5f-b81a-41c8-beca-f0b9d598aae4","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Learning to reason with llms","work_id":"16f7723e-0b48-4125-885f-310de5bf2c28","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Gemini 2.0 flash thinking, 2024","work_id":"ae6b4683-11f7-4205-91f3-a116f2c49f43","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Grok 3 beta — the age of reasoning agents, 2024","work_id":"21963273-aa23-44a9-be08-9bd3fd2bc191","year":2024}],"snapshot_sha256":"ab37c378c53f205646ae8c52b2abcc0c15c8085d63d5d0d0c88ec0179c9d0268"},"source":{"id":"2507.02259","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T11:11:50.392302Z","id":"49cf8c9c-1fb9-4c13-b384-d5862d5733c3","model_set":{"reader":"grok-4.3"},"one_line_summary":"MemAgent uses multi-conversation RL to train a memory agent that reads text in segments and overwrites memory, extrapolating from 8K training to 3.5M token QA with under 5% loss and 95%+ on 512K RULER.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"MemAgent lets LLMs handle millions of tokens by segmenting input and overwriting memory after RL training on 32K texts.","strongest_claim":"MemAgent has demonstrated superb long-context capabilities, being able to extrapolate from an 8K context trained on 32K text to a 3.5M QA task with performance loss < 5% and achieves 95%+ in 512K RULER test.","weakest_assumption":"That the overwrite memory strategy combined with multi-conversation RL training will continue to prevent performance degradation when scaling far beyond the 32K training length without additional mechanisms or data."}},"verdict_id":"49cf8c9c-1fb9-4c13-b384-d5862d5733c3"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:739808455c268cb3ea5453277c271035686f81361fd941883882b4eb94b24269","target":"record","created_at":"2026-05-17T23:38:52Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"909f67bdf57361e1e01dbde35b145d65f5a3e5e6d9083acc5b7b887a65a2b28d","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-03T03:11:50Z","title_canon_sha256":"1c5f2eb4f7d03e2225f06250d7bc510eab14efdecea0d98d161a35291af95019"},"schema_version":"1.0","source":{"id":"2507.02259","kind":"arxiv","version":1}},"canonical_sha256":"d8144c937303d083e702ec438fe467cb66da96ccf76f3bf79fd0a307195bf4b8","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"d8144c937303d083e702ec438fe467cb66da96ccf76f3bf79fd0a307195bf4b8","first_computed_at":"2026-05-17T23:38:52.698861Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:52.698861Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"XWNAIYQaeb5zNw5Lm6Ln3O2q/+WK3vJh4XaAF4oRqu6xlGwnpkVn5JmzUAQn4m2hXsTUFDePrqO7DRuE9752DQ==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:52.699547Z","signed_message":"canonical_sha256_bytes"},"source_id":"2507.02259","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:739808455c268cb3ea5453277c271035686f81361fd941883882b4eb94b24269","sha256:a4bb4eb0b70b7a614dd1b81e215a99b5597d42617e9b4d41be2caa3057e83e7b"],"state_sha256":"7a96431d6154b9b80c1abd3947a6b8e5d1af62367e7b9ffc61f9657d27d31957"}