{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:I452AMLSZVTA4RBYR335OPRJRE","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":"ad633a244a79a763073ea1df51bc22044cae72f9b828905346cab0a22cbf0868","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-23T17:55:31Z","title_canon_sha256":"021e9c596b91e444484c14a187a7d3dc84869cdd4b13f68ee4532b1dffe16db7"},"schema_version":"1.0","source":{"id":"2310.15154","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2310.15154","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"arxiv_version","alias_value":"2310.15154v1","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.15154","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"pith_short_12","alias_value":"I452AMLSZVTA","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"I452AMLSZVTA4RBY","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"I452AMLS","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:ebf47cff60fd92a65fe63e0bbbc5dc652d754159addc49ccc254ce6319ec9942","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":"sentiment is represented linearly: a single direction in activation space mostly captures the feature across a range of tasks with one extreme for positive and the other for negative."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the identified direction is the primary and stable representation of sentiment rather than one of several correlated directions that happen to align on the chosen datasets and models."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Sentiment is represented as a single linear direction in LLM activation space that is causally relevant across tasks and is summarized at punctuation and names in addition to charged words."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Sentiment in large language models is captured by one direction in activation space, with positive and negative at opposite poles."}],"snapshot_sha256":"723ea1bd44ac9b4ad91a7f845cf064cb9e939837ad4f7cde5bc3caa21bd41d40"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"06ce9b182fafc8bb3d45d9a56cc920cce5d23b4fa4ded86a9e6959511795d9f9"},"paper":{"abstract_excerpt":"Sentiment is a pervasive feature in natural language text, yet it is an open question how sentiment is represented within Large Language Models (LLMs). In this study, we reveal that across a range of models, sentiment is represented linearly: a single direction in activation space mostly captures the feature across a range of tasks with one extreme for positive and the other for negative. Through causal interventions, we isolate this direction and show it is causally relevant in both toy tasks and real world datasets such as Stanford Sentiment Treebank. Through this case study we model a thoro","authors_text":"Atticus Geiger, Curt Tigges, Neel Nanda, Oskar John Hollinsworth","cross_cats":["cs.AI","cs.CL"],"headline":"Sentiment in large language models is captured by one direction in activation space, with positive and negative at opposite poles.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-23T17:55:31Z","title":"Linear Representations of Sentiment in Large Language Models"},"references":{"count":122,"internal_anchors":1,"resolved_work":122,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Eliciting latent knowledge: How to tell if your eyes deceive you , author=. 2021 , month=","work_id":"97920acf-6383-4636-b93b-cf646b7b9e8f","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Karl Pearson F.R.S. , title =. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science , volume =. 1901 , publisher =","work_id":"b8ab6ba1-6f30-436e-8feb-8eaa39326473","year":1901},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Information Theory, IEEE Transactions on , volume=","work_id":"87b0e008-4246-4fce-9c7c-fd4b36790357","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Journal of the Royal Statistical Society: Series B (Methodological) , volume=","work_id":"7e058df2-2b5c-46e8-b8f1-a584271c6e0b","year":1958},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Finding Alignments Between Interpretable Causal Variables and Distributed Neural Representations , author=. 2023 , eprint=","work_id":"cbb90bdc-7161-4182-a54f-aa9b25aaf24a","year":2023}],"snapshot_sha256":"8808a5fa2619ec3cd2a54608ce5291a941fe210471ba586ef8acd348a1f0a3ad"},"source":{"id":"2310.15154","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T12:37:30.988087Z","id":"fa3b0018-a569-4c41-a321-403d043b76fd","model_set":{"reader":"grok-4.3"},"one_line_summary":"Sentiment is represented as a single linear direction in LLM activation space that is causally relevant across tasks and is summarized at punctuation and names in addition to charged words.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Sentiment in large language models is captured by one direction in activation space, with positive and negative at opposite poles.","strongest_claim":"sentiment is represented linearly: a single direction in activation space mostly captures the feature across a range of tasks with one extreme for positive and the other for negative.","weakest_assumption":"That the identified direction is the primary and stable representation of sentiment rather than one of several correlated directions that happen to align on the chosen datasets and models."}},"verdict_id":"fa3b0018-a569-4c41-a321-403d043b76fd"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:58122707ad20fa156890c7789c6c77ff52fdf88a1452632b06b27b7d5ec4b7b7","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":"ad633a244a79a763073ea1df51bc22044cae72f9b828905346cab0a22cbf0868","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-23T17:55:31Z","title_canon_sha256":"021e9c596b91e444484c14a187a7d3dc84869cdd4b13f68ee4532b1dffe16db7"},"schema_version":"1.0","source":{"id":"2310.15154","kind":"arxiv","version":1}},"canonical_sha256":"473ba03172cd660e44388ef7d73e29892ff73fd0832dd623106b5271a5755b36","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"473ba03172cd660e44388ef7d73e29892ff73fd0832dd623106b5271a5755b36","first_computed_at":"2026-05-17T23:38:52.529178Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:52.529178Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"0AxnbSg0saqboJbW1z/Qt0/IuJ/LT+ywshKVeQoRTVT69V6+UIUrOepHUeFp2eqkeaDfcZAIipAZDittSed/Bw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:52.529627Z","signed_message":"canonical_sha256_bytes"},"source_id":"2310.15154","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:58122707ad20fa156890c7789c6c77ff52fdf88a1452632b06b27b7d5ec4b7b7","sha256:ebf47cff60fd92a65fe63e0bbbc5dc652d754159addc49ccc254ce6319ec9942"],"state_sha256":"8cd7743eb59079b9e26bbf15586ed5dc55a6b782e29d4545be4423db1db7f6ca"}