{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CJLXO7D455YHTEUUZU56XPIZTE","short_pith_number":"pith:CJLXO7D4","schema_version":"1.0","canonical_sha256":"1257777c7cef70799294cd3bebbd1999314cf841642e7c34d4bd402e5049d209","source":{"kind":"arxiv","id":"2410.01280","version":1},"attestation_state":"computed","paper":{"title":"Sparse Autoencoders Reveal Temporal Difference Learning in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Akshay K. Jagadish, Can Demircan, Eric Schulz, Marcel Binz, Tankred Saanum","submitted_at":"2024-10-02T06:51:12Z","abstract_excerpt":"In-context learning, the ability to adapt based on a few examples in the input prompt, is a ubiquitous feature of large language models (LLMs). However, as LLMs' in-context learning abilities continue to improve, understanding this phenomenon mechanistically becomes increasingly important. In particular, it is not well-understood how LLMs learn to solve specific classes of problems, such as reinforcement learning (RL) problems, in-context. Through three different tasks, we first show that Llama $3$ $70$B can solve simple RL problems in-context. We then analyze the residual stream of Llama usin"},"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":"2410.01280","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-02T06:51:12Z","cross_cats_sorted":[],"title_canon_sha256":"51476210ff81ff8d5caaa7b5dbf0ede6bbcc7304dcae0cd2c073343f3d6e9fa5","abstract_canon_sha256":"895a3754a864ea133f21f64c2fbd2f199b0f3fa7d85e36a743ce8da0b05129c2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:14:45.711115Z","signature_b64":"lsh/RWGNsXLCpNpAlR0bbpW92e6Vu89wXwdU0KghVdAI6RW6ePpoxbk+F7cvLOCXeUmJ38z6geRp3bRtzVQJCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1257777c7cef70799294cd3bebbd1999314cf841642e7c34d4bd402e5049d209","last_reissued_at":"2026-07-05T09:14:45.710655Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:14:45.710655Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sparse Autoencoders Reveal Temporal Difference Learning in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Akshay K. Jagadish, Can Demircan, Eric Schulz, Marcel Binz, Tankred Saanum","submitted_at":"2024-10-02T06:51:12Z","abstract_excerpt":"In-context learning, the ability to adapt based on a few examples in the input prompt, is a ubiquitous feature of large language models (LLMs). However, as LLMs' in-context learning abilities continue to improve, understanding this phenomenon mechanistically becomes increasingly important. In particular, it is not well-understood how LLMs learn to solve specific classes of problems, such as reinforcement learning (RL) problems, in-context. Through three different tasks, we first show that Llama $3$ $70$B can solve simple RL problems in-context. We then analyze the residual stream of Llama usin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.01280","kind":"arxiv","version":1},"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/2410.01280/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":"2410.01280","created_at":"2026-07-05T09:14:45.710716+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.01280v1","created_at":"2026-07-05T09:14:45.710716+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.01280","created_at":"2026-07-05T09:14:45.710716+00:00"},{"alias_kind":"pith_short_12","alias_value":"CJLXO7D455YH","created_at":"2026-07-05T09:14:45.710716+00:00"},{"alias_kind":"pith_short_16","alias_value":"CJLXO7D455YHTEUU","created_at":"2026-07-05T09:14:45.710716+00:00"},{"alias_kind":"pith_short_8","alias_value":"CJLXO7D4","created_at":"2026-07-05T09:14:45.710716+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23040","citing_title":"Steered Generation via Gradient-Based Optimization on Sparse Query Features","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12412","citing_title":"Stories in Space: In-Context Learning Trajectories in Conceptual Belief Space","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05115","citing_title":"Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior","ref_index":260,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE","json":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE.json","graph_json":"https://pith.science/api/pith-number/CJLXO7D455YHTEUUZU56XPIZTE/graph.json","events_json":"https://pith.science/api/pith-number/CJLXO7D455YHTEUUZU56XPIZTE/events.json","paper":"https://pith.science/paper/CJLXO7D4"},"agent_actions":{"view_html":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE","download_json":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE.json","view_paper":"https://pith.science/paper/CJLXO7D4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.01280&json=true","fetch_graph":"https://pith.science/api/pith-number/CJLXO7D455YHTEUUZU56XPIZTE/graph.json","fetch_events":"https://pith.science/api/pith-number/CJLXO7D455YHTEUUZU56XPIZTE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE/action/storage_attestation","attest_author":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE/action/author_attestation","sign_citation":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE/action/citation_signature","submit_replication":"https://pith.science/pith/CJLXO7D455YHTEUUZU56XPIZTE/action/replication_record"}},"created_at":"2026-07-05T09:14:45.710716+00:00","updated_at":"2026-07-05T09:14:45.710716+00:00"}