{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:A3URAZXM4HTPHU3KZ4WXLMH7N7","short_pith_number":"pith:A3URAZXM","canonical_record":{"source":{"id":"2507.06427","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-08T22:17:52Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"e81bc752d5b9c8fa4222e80ac90a72648794d272294e5f48b8ea9783f3f218e3","abstract_canon_sha256":"cc3b9e0031ce0a74da0d24e049ab35107762595b718038e66239729967ed3361"},"schema_version":"1.0"},"canonical_sha256":"06e91066ece1e6f3d36acf2d75b0ff6fcb465320dd06766443cfbfd34e4c41c4","source":{"kind":"arxiv","id":"2507.06427","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2507.06427","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"arxiv_version","alias_value":"2507.06427v1","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.06427","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"pith_short_12","alias_value":"A3URAZXM4HTP","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"pith_short_16","alias_value":"A3URAZXM4HTPHU3K","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"pith_short_8","alias_value":"A3URAZXM","created_at":"2026-07-05T11:34:15Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:A3URAZXM4HTPHU3KZ4WXLMH7N7","target":"record","payload":{"canonical_record":{"source":{"id":"2507.06427","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-08T22:17:52Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"e81bc752d5b9c8fa4222e80ac90a72648794d272294e5f48b8ea9783f3f218e3","abstract_canon_sha256":"cc3b9e0031ce0a74da0d24e049ab35107762595b718038e66239729967ed3361"},"schema_version":"1.0"},"canonical_sha256":"06e91066ece1e6f3d36acf2d75b0ff6fcb465320dd06766443cfbfd34e4c41c4","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:34:15.607131Z","signature_b64":"Kt1wZl+mvL9ZImwNxPvhWC+xGUyqWomvfmUe+ec+1UdIK5i7lgyqVRGNrx1lqUWVaZklbRK4nHcO41IVEYAlCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"06e91066ece1e6f3d36acf2d75b0ff6fcb465320dd06766443cfbfd34e4c41c4","last_reissued_at":"2026-07-05T11:34:15.606672Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:34:15.606672Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2507.06427","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:34:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"s36kAGMyrdS4vMBMXo9JujydK9gz9d8JF9tjxBhh1Ldd4VC/LaZtHu1261uHNx6yeloVSVxpXXLN3vi+AE4aCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T02:35:22.054444Z"},"content_sha256":"c1bb8ae972f45c7805ae1fce32558b49f2d31afa92f8049e05a33231d836a01e","schema_version":"1.0","event_id":"sha256:c1bb8ae972f45c7805ae1fce32558b49f2d31afa92f8049e05a33231d836a01e"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:A3URAZXM4HTPHU3KZ4WXLMH7N7","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Exploring Task Performance with Interpretable Models via Sparse Auto-Encoders","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bohao Yang, Chenghua Lin, Dong Yang, Shun Wang, Tyler Loakman, Yi Liu, Youbo Lei, Yuting Zhao","submitted_at":"2025-07-08T22:17:52Z","abstract_excerpt":"Large Language Models (LLMs) are traditionally viewed as black-box algorithms, therefore reducing trustworthiness and obscuring potential approaches to increasing performance on downstream tasks. In this work, we apply an effective LLM decomposition method using a dictionary-learning approach with sparse autoencoders. This helps extract monosemantic features from polysemantic LLM neurons. Remarkably, our work identifies model-internal misunderstanding, allowing the automatic reformulation of the prompts with additional annotations to improve the interpretation by LLMs. Moreover, this approach "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.06427","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/2507.06427/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:34:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"RKwBNsFq44PPPyctk4k+sO4dGgovPbGerkiWfVuAZBXxYs4Zne1hZ++fbdf+tg0PraIjKsqc1Zo+H4p7L33DBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T02:35:22.054946Z"},"content_sha256":"73cc4b15e3ab77bb0d608b33a0b1201e5584e72ecc9a1e4e0f1d151ed1bd43bb","schema_version":"1.0","event_id":"sha256:73cc4b15e3ab77bb0d608b33a0b1201e5584e72ecc9a1e4e0f1d151ed1bd43bb"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/A3URAZXM4HTPHU3KZ4WXLMH7N7/bundle.json","state_url":"https://pith.science/pith/A3URAZXM4HTPHU3KZ4WXLMH7N7/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/A3URAZXM4HTPHU3KZ4WXLMH7N7/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-08T02:35:22Z","links":{"resolver":"https://pith.science/pith/A3URAZXM4HTPHU3KZ4WXLMH7N7","bundle":"https://pith.science/pith/A3URAZXM4HTPHU3KZ4WXLMH7N7/bundle.json","state":"https://pith.science/pith/A3URAZXM4HTPHU3KZ4WXLMH7N7/state.json","well_known_bundle":"https://pith.science/.well-known/pith/A3URAZXM4HTPHU3KZ4WXLMH7N7/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:A3URAZXM4HTPHU3KZ4WXLMH7N7","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":"cc3b9e0031ce0a74da0d24e049ab35107762595b718038e66239729967ed3361","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-08T22:17:52Z","title_canon_sha256":"e81bc752d5b9c8fa4222e80ac90a72648794d272294e5f48b8ea9783f3f218e3"},"schema_version":"1.0","source":{"id":"2507.06427","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2507.06427","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"arxiv_version","alias_value":"2507.06427v1","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.06427","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"pith_short_12","alias_value":"A3URAZXM4HTP","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"pith_short_16","alias_value":"A3URAZXM4HTPHU3K","created_at":"2026-07-05T11:34:15Z"},{"alias_kind":"pith_short_8","alias_value":"A3URAZXM","created_at":"2026-07-05T11:34:15Z"}],"graph_snapshots":[{"event_id":"sha256:73cc4b15e3ab77bb0d608b33a0b1201e5584e72ecc9a1e4e0f1d151ed1bd43bb","target":"graph","created_at":"2026-07-05T11:34:15Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2507.06427/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large Language Models (LLMs) are traditionally viewed as black-box algorithms, therefore reducing trustworthiness and obscuring potential approaches to increasing performance on downstream tasks. In this work, we apply an effective LLM decomposition method using a dictionary-learning approach with sparse autoencoders. This helps extract monosemantic features from polysemantic LLM neurons. Remarkably, our work identifies model-internal misunderstanding, allowing the automatic reformulation of the prompts with additional annotations to improve the interpretation by LLMs. Moreover, this approach ","authors_text":"Bohao Yang, Chenghua Lin, Dong Yang, Shun Wang, Tyler Loakman, Yi Liu, Youbo Lei, Yuting Zhao","cross_cats":["cs.LG"],"headline":"","license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-08T22:17:52Z","title":"Exploring Task Performance with Interpretable Models via Sparse Auto-Encoders"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.06427","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c1bb8ae972f45c7805ae1fce32558b49f2d31afa92f8049e05a33231d836a01e","target":"record","created_at":"2026-07-05T11:34:15Z","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":"cc3b9e0031ce0a74da0d24e049ab35107762595b718038e66239729967ed3361","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-08T22:17:52Z","title_canon_sha256":"e81bc752d5b9c8fa4222e80ac90a72648794d272294e5f48b8ea9783f3f218e3"},"schema_version":"1.0","source":{"id":"2507.06427","kind":"arxiv","version":1}},"canonical_sha256":"06e91066ece1e6f3d36acf2d75b0ff6fcb465320dd06766443cfbfd34e4c41c4","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"06e91066ece1e6f3d36acf2d75b0ff6fcb465320dd06766443cfbfd34e4c41c4","first_computed_at":"2026-07-05T11:34:15.606672Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:34:15.606672Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Kt1wZl+mvL9ZImwNxPvhWC+xGUyqWomvfmUe+ec+1UdIK5i7lgyqVRGNrx1lqUWVaZklbRK4nHcO41IVEYAlCQ==","signature_status":"signed_v1","signed_at":"2026-07-05T11:34:15.607131Z","signed_message":"canonical_sha256_bytes"},"source_id":"2507.06427","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c1bb8ae972f45c7805ae1fce32558b49f2d31afa92f8049e05a33231d836a01e","sha256:73cc4b15e3ab77bb0d608b33a0b1201e5584e72ecc9a1e4e0f1d151ed1bd43bb"],"state_sha256":"4a798de76a589f9ccc2fea0d2b729335af6c7a985e3ce2fc506e1cee6188ea81"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"mGD/UJwKhmmtGvRFMAFZBbSdyS5wE9jlzFvZ/Pb2S5QgjHaFUU3WWBtUgKEcQeKla/XzJLUX2fB81CtEZz8NAw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-08T02:35:22.059732Z","bundle_sha256":"f4f1fd74203789f1f135311b9d805ec77427b1cb1c3b5d412e36c464e9ae9873"}}