{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:2Q6SC5MD42YNMEH5S4GWFMKH2P","short_pith_number":"pith:2Q6SC5MD","schema_version":"1.0","canonical_sha256":"d43d217583e6b0d610fd970d62b147d3c865ecaf1971c467935fe431a68e1b72","source":{"kind":"arxiv","id":"2505.18235","version":1},"attestation_state":"computed","paper":{"title":"The Origins of Representation Manifolds in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alexander Modell, Nick Whiteley, Patrick Rubin-Delanchy","submitted_at":"2025-05-23T13:31:22Z","abstract_excerpt":"There is a large ongoing scientific effort in mechanistic interpretability to map embeddings and internal representations of AI systems into human-understandable concepts. A key element of this effort is the linear representation hypothesis, which posits that neural representations are sparse linear combinations of `almost-orthogonal' direction vectors, reflecting the presence or absence of different features. This model underpins the use of sparse autoencoders to recover features from representations. Moving towards a fuller model of features, in which neural representations could encode not "},"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":"2505.18235","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-23T13:31:22Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"44b792dd7775ffaa0d13ba95cb1ac71f696ab79e52c3d5656b5155414ff6a20f","abstract_canon_sha256":"f281802cb3f41bb3d70a94dcb1f542f94285701a14064b3b30f30863f48f3dc1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:08:43.251107Z","signature_b64":"nSrYG0W8T35DrU7jDniFby46OcqzPu25SIO2JsXdtwzSwf2NGHHV9I+gk7Yu8kkaK/YrBm1QQMtBNtLjnsQ8Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d43d217583e6b0d610fd970d62b147d3c865ecaf1971c467935fe431a68e1b72","last_reissued_at":"2026-07-05T11:08:43.250576Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:08:43.250576Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Origins of Representation Manifolds in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alexander Modell, Nick Whiteley, Patrick Rubin-Delanchy","submitted_at":"2025-05-23T13:31:22Z","abstract_excerpt":"There is a large ongoing scientific effort in mechanistic interpretability to map embeddings and internal representations of AI systems into human-understandable concepts. A key element of this effort is the linear representation hypothesis, which posits that neural representations are sparse linear combinations of `almost-orthogonal' direction vectors, reflecting the presence or absence of different features. This model underpins the use of sparse autoencoders to recover features from representations. Moving towards a fuller model of features, in which neural representations could encode not "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.18235","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/2505.18235/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":"2505.18235","created_at":"2026-07-05T11:08:43.250634+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.18235v1","created_at":"2026-07-05T11:08:43.250634+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.18235","created_at":"2026-07-05T11:08:43.250634+00:00"},{"alias_kind":"pith_short_12","alias_value":"2Q6SC5MD42YN","created_at":"2026-07-05T11:08:43.250634+00:00"},{"alias_kind":"pith_short_16","alias_value":"2Q6SC5MD42YNMEH5","created_at":"2026-07-05T11:08:43.250634+00:00"},{"alias_kind":"pith_short_8","alias_value":"2Q6SC5MD","created_at":"2026-07-05T11:08:43.250634+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25234","citing_title":"Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifolds","ref_index":83,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26361","citing_title":"Does Aurora Encode Atmospheric Structure? Latent Regime Analysis and Attribution","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01774","citing_title":"Subliminal Clocks: Latent Time Modelling in Diffusion Language Models","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09894","citing_title":"A Navigable Manifold of Hypothesized Consciousness-Spectrum States in Language Model Representations","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05315","citing_title":"LoRi: Low-Rank Distillation for Implicit Reasoning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05194","citing_title":"Temporal Preference Concepts and their Functions in a Large Language Model","ref_index":77,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01844","citing_title":"The Cylindrical Representation Hypothesis for Language Model Steering","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21706","citing_title":"Latent-space Attacks for Refusal Evasion in Language Models","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29126","citing_title":"When and How Long? The Readout-Mediator Angle in Temporal Reasoning","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06333","citing_title":"Subspace-Aware Sparse Autoencoders for Effective Mechanistic Interpretability","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26396","citing_title":"At the Edge of Understanding: Sparse Autoencoders Trace The Limits of Transformer Generalization","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23821","citing_title":"Hierarchical Concept Geometry in Language Models Emerges from Word Co-occurrence","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21706","citing_title":"Latent-space Attacks for Refusal Evasion in Language Models","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18537","citing_title":"Probing for Representation Manifolds in Superposition","ref_index":97,"is_internal_anchor":false},{"citing_arxiv_id":"2505.20340","citing_title":"Latent Trajectory Dynamics in Large Language Models: A Manifold Evolution Framework with Empirical Validation","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14075","citing_title":"Rethinking Layer Relevance in Large Language Models Beyond Cosine Similarity","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09967","citing_title":"Tensor Product Representation Probes Reveal Shared Structure Across Linear Directions","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09224","citing_title":"SMIXAE: Towards Unsupervised Manifold Discovery in Language Models","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P","json":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P.json","graph_json":"https://pith.science/api/pith-number/2Q6SC5MD42YNMEH5S4GWFMKH2P/graph.json","events_json":"https://pith.science/api/pith-number/2Q6SC5MD42YNMEH5S4GWFMKH2P/events.json","paper":"https://pith.science/paper/2Q6SC5MD"},"agent_actions":{"view_html":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P","download_json":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P.json","view_paper":"https://pith.science/paper/2Q6SC5MD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.18235&json=true","fetch_graph":"https://pith.science/api/pith-number/2Q6SC5MD42YNMEH5S4GWFMKH2P/graph.json","fetch_events":"https://pith.science/api/pith-number/2Q6SC5MD42YNMEH5S4GWFMKH2P/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P/action/storage_attestation","attest_author":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P/action/author_attestation","sign_citation":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P/action/citation_signature","submit_replication":"https://pith.science/pith/2Q6SC5MD42YNMEH5S4GWFMKH2P/action/replication_record"}},"created_at":"2026-07-05T11:08:43.250634+00:00","updated_at":"2026-07-05T11:08:43.250634+00:00"}