{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:D7YQCPCHYYKWIV7BJEURZPN6VH","short_pith_number":"pith:D7YQCPCH","schema_version":"1.0","canonical_sha256":"1ff1013c47c6156457e149291cbdbea9e98558b800fb1ad00c698db0df2b4a57","source":{"kind":"arxiv","id":"2602.10520","version":3},"attestation_state":"computed","paper":{"title":"Prioritize the Process, Not Just the Outcome: Rewarding Latent Thought Trajectories Improves Reasoning in Looped Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Esin Tureci, Jonathan Williams","submitted_at":"2026-02-11T04:39:42Z","abstract_excerpt":"Looped Language Models (LoopLMs) perform multi-step latent reasoning prior to token generation and outperform conventional LLMs on reasoning benchmarks at smaller parameter budgets. However, attempts to further improve LoopLM reasoning with reinforcement learning have failed - standard objectives such as Group Relative Policy Optimization (GRPO) only assign credit to the final latent state, creating a fundamental mismatch with the model's internal computation. To resolve this, we introduce RLTT (Reward Latent Thought Trajectories), a reinforcement learning framework which distributes reward ac"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2602.10520","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-02-11T04:39:42Z","cross_cats_sorted":[],"title_canon_sha256":"931a8c64075c56d2b425397d3b2b0516de4464167550275060031f42db9dd5cf","abstract_canon_sha256":"08e1903f20e847ecb264d3e6ee352009a3a464e345bfa72199b20ee65966e805"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T02:05:02.115621Z","signature_b64":"Gll3cT3ng+yLEcxzhiDK2sNKiZ8UyXeQQcmoentKEDqVKQxag5zZWdXj+xuxB6tbzd3BBXxBvJki2457BZwiBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1ff1013c47c6156457e149291cbdbea9e98558b800fb1ad00c698db0df2b4a57","last_reissued_at":"2026-05-29T02:05:02.114788Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T02:05:02.114788Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Prioritize the Process, Not Just the Outcome: Rewarding Latent Thought Trajectories Improves Reasoning in Looped Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Esin Tureci, Jonathan Williams","submitted_at":"2026-02-11T04:39:42Z","abstract_excerpt":"Looped Language Models (LoopLMs) perform multi-step latent reasoning prior to token generation and outperform conventional LLMs on reasoning benchmarks at smaller parameter budgets. However, attempts to further improve LoopLM reasoning with reinforcement learning have failed - standard objectives such as Group Relative Policy Optimization (GRPO) only assign credit to the final latent state, creating a fundamental mismatch with the model's internal computation. To resolve this, we introduce RLTT (Reward Latent Thought Trajectories), a reinforcement learning framework which distributes reward ac"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.10520","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.10520/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":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2602.10520","created_at":"2026-05-29T02:05:02.114892+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.10520v3","created_at":"2026-05-29T02:05:02.114892+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.10520","created_at":"2026-05-29T02:05:02.114892+00:00"},{"alias_kind":"pith_short_12","alias_value":"D7YQCPCHYYKW","created_at":"2026-05-29T02:05:02.114892+00:00"},{"alias_kind":"pith_short_16","alias_value":"D7YQCPCHYYKWIV7B","created_at":"2026-05-29T02:05:02.114892+00:00"},{"alias_kind":"pith_short_8","alias_value":"D7YQCPCH","created_at":"2026-05-29T02:05:02.114892+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.29712","citing_title":"Why Struggle with Continuous Latents? Interpretable Discrete Latent Reasoning via Rendered Compression","ref_index":49,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH","json":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH.json","graph_json":"https://pith.science/api/pith-number/D7YQCPCHYYKWIV7BJEURZPN6VH/graph.json","events_json":"https://pith.science/api/pith-number/D7YQCPCHYYKWIV7BJEURZPN6VH/events.json","paper":"https://pith.science/paper/D7YQCPCH"},"agent_actions":{"view_html":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH","download_json":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH.json","view_paper":"https://pith.science/paper/D7YQCPCH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.10520&json=true","fetch_graph":"https://pith.science/api/pith-number/D7YQCPCHYYKWIV7BJEURZPN6VH/graph.json","fetch_events":"https://pith.science/api/pith-number/D7YQCPCHYYKWIV7BJEURZPN6VH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH/action/storage_attestation","attest_author":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH/action/author_attestation","sign_citation":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH/action/citation_signature","submit_replication":"https://pith.science/pith/D7YQCPCHYYKWIV7BJEURZPN6VH/action/replication_record"}},"created_at":"2026-05-29T02:05:02.114892+00:00","updated_at":"2026-05-29T02:05:02.114892+00:00"}