{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:QL3Z3XGWSFS7EUJW2GDLTBLYCE","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":"71d6037dba20b7f7c55b912ad20faab32edd0e42637bb833153811c7edbff7b3","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T23:01:29Z","title_canon_sha256":"1ded6d8f41fe302a9f70947be501078be8aa2e08c6c05a782ffd9f0797aff66a"},"schema_version":"1.0","source":{"id":"2605.12809","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12809","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12809v1","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12809","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"pith_short_12","alias_value":"QL3Z3XGWSFS7","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"QL3Z3XGWSFS7EUJW","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"QL3Z3XGW","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:7186b25db16fce8107af569a8f1e2ec0eb7d2fc8ca0f6a884e63700ed7f0432b","target":"graph","created_at":"2026-05-18T03:09:12Z","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":"We introduce a flexible framework that infers token-level influence through a latent mediation approach for general prediction tasks... Token-level influence is obtained by propagating latent attributions back to the input space via token activation patterns."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The assumption that sparse autoencoders attached to LLM layers learn a basis of approximately independent latent features whose influence can be accurately propagated via Jacobian-vector products without introducing new biases."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"A latent mediation framework with sparse autoencoders enables non-additive token-level influence attribution in LLMs by learning orthogonal features and back-propagating attributions."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A latent mediation method using sparse autoencoders delivers reliable token-level influence attribution for LLM predictions on any task."}],"snapshot_sha256":"35dbf689b3acf125df8d4bb6ccbb9ef6005c9f1f5168ab2d2995135279e05403"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"A critical step for reliable large language models (LLMs) use in healthcare is to attribute predictions to their training data, akin to a medical case study. This requires token-level precision: pinpointing not just which training examples influence a decision, but which tokens within them are responsible. While influence functions offer a principled framework for this, prior work is restricted to autoregressive settings and relies on an implicit assumption of token independence, rendering their identified influences unreliable. We introduce a flexible framework that infers token-level influen","authors_text":"Kyra Gan, Promit Ghosal, Shixing Yu","cross_cats":["cs.AI"],"headline":"A latent mediation method using sparse autoencoders delivers reliable token-level influence attribution for LLM predictions on any task.","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T23:01:29Z","title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces"},"references":{"count":300,"internal_anchors":50,"resolved_work":300,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"International Conference on Learning Representations , year=","work_id":"f975b9b8-13c7-4593-9830-1887eee9d305","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume=","work_id":"dd6121d7-f0e7-43ad-9da1-25608f600c48","year":2013},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Philosophical transactions of the royal society A: Mathematical, Physical and Engineering Sciences , volume=","work_id":"41da284d-638e-40b8-bf3e-19af20c218ed","year":2016},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"arXiv preprint arXiv:2411.07618 , year=","work_id":"c0e4d0c8-7272-40d3-83e2-44de8b252f9f","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Transformer Circuits Thread , volume=","work_id":"bb66d71e-4735-49aa-af5f-610b0164a701","year":null}],"snapshot_sha256":"147f2defe4e18329597fa089ffd4c2c7fe5e3b8ef3f87dc2d21434a56197bc3f"},"source":{"id":"2605.12809","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:14:06.178610Z","id":"d614feb7-4442-45a3-bb37-9da913b1d86f","model_set":{"reader":"grok-4.3"},"one_line_summary":"A latent mediation framework with sparse autoencoders enables non-additive token-level influence attribution in LLMs by learning orthogonal features and back-propagating attributions.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A latent mediation method using sparse autoencoders delivers reliable token-level influence attribution for LLM predictions on any task.","strongest_claim":"We introduce a flexible framework that infers token-level influence through a latent mediation approach for general prediction tasks... Token-level influence is obtained by propagating latent attributions back to the input space via token activation patterns.","weakest_assumption":"The assumption that sparse autoencoders attached to LLM layers learn a basis of approximately independent latent features whose influence can be accurately propagated via Jacobian-vector products without introducing new biases."}},"verdict_id":"d614feb7-4442-45a3-bb37-9da913b1d86f"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:134f215acbfdc98deb2f1621ca371f839c86b66f88333a2f2b5679008253c317","target":"record","created_at":"2026-05-18T03:09:12Z","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":"71d6037dba20b7f7c55b912ad20faab32edd0e42637bb833153811c7edbff7b3","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T23:01:29Z","title_canon_sha256":"1ded6d8f41fe302a9f70947be501078be8aa2e08c6c05a782ffd9f0797aff66a"},"schema_version":"1.0","source":{"id":"2605.12809","kind":"arxiv","version":1}},"canonical_sha256":"82f79ddcd69165f25136d186b9857811118d31d9f06f20b1128a3afa44afe3fd","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"82f79ddcd69165f25136d186b9857811118d31d9f06f20b1128a3afa44afe3fd","first_computed_at":"2026-05-18T03:09:12.536473Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:12.536473Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"TMYte6YY94xGDIj0jClfqBIQgH6PuSkgEiCRdjKHKPADkLVg8Y7flvlu9g42Le2CV9Le0r9GJJyIGlIq0wHzBw==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:12.537487Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12809","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:134f215acbfdc98deb2f1621ca371f839c86b66f88333a2f2b5679008253c317","sha256:7186b25db16fce8107af569a8f1e2ec0eb7d2fc8ca0f6a884e63700ed7f0432b"],"state_sha256":"26bc304a8d609f8dca3dfb323b7547096c8c7d0d54643966ba028717bc10e668"}