{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:V7XL7TQMKV4RUZMMBSCWNLMFFN","short_pith_number":"pith:V7XL7TQM","schema_version":"1.0","canonical_sha256":"afeebfce0c55791a658c0c8566ad852b53417e1ccdb7759610159efdca500785","source":{"kind":"arxiv","id":"2504.16318","version":2},"attestation_state":"computed","paper":{"title":"Semantics at an Angle: When Cosine Similarity Works Until It Doesn't","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Kisung You","submitted_at":"2025-04-22T23:31:32Z","abstract_excerpt":"Cosine similarity has become a standard metric for comparing embeddings in modern machine learning. Its scale-invariance and alignment with model training objectives have contributed to its widespread adoption. However, recent studies have revealed important limitations, particularly when embedding norms carry meaningful semantic information. This informal article offers a reflective and selective examination of the evolution, strengths, and limitations of cosine similarity. We highlight why it performs well in many settings, where it tends to break down, and how emerging alternatives are begi"},"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":"2504.16318","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-22T23:31:32Z","cross_cats_sorted":[],"title_canon_sha256":"a6cea29ba332cae088fcf6f558e7aae43f01052fdbfaeee6bea696e287e6a01d","abstract_canon_sha256":"bd4d5ef562bdf4ad6adc85ba683c4aea6fe04e752248252e0d0ea2060ed02c7c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:06:13.252508Z","signature_b64":"rfJFOgDnhq8CPsVzEhVOrZGNnXgfMJL1CfY45+dWTahAqKE7EML+dT7Te6b9ca0Cd6psoRodOplK/RYNi9fnDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"afeebfce0c55791a658c0c8566ad852b53417e1ccdb7759610159efdca500785","last_reissued_at":"2026-07-05T11:06:13.251942Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:06:13.251942Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Semantics at an Angle: When Cosine Similarity Works Until It Doesn't","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Kisung You","submitted_at":"2025-04-22T23:31:32Z","abstract_excerpt":"Cosine similarity has become a standard metric for comparing embeddings in modern machine learning. Its scale-invariance and alignment with model training objectives have contributed to its widespread adoption. However, recent studies have revealed important limitations, particularly when embedding norms carry meaningful semantic information. This informal article offers a reflective and selective examination of the evolution, strengths, and limitations of cosine similarity. We highlight why it performs well in many settings, where it tends to break down, and how emerging alternatives are begi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.16318","kind":"arxiv","version":2},"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/2504.16318/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":"2504.16318","created_at":"2026-07-05T11:06:13.252000+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.16318v2","created_at":"2026-07-05T11:06:13.252000+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.16318","created_at":"2026-07-05T11:06:13.252000+00:00"},{"alias_kind":"pith_short_12","alias_value":"V7XL7TQMKV4R","created_at":"2026-07-05T11:06:13.252000+00:00"},{"alias_kind":"pith_short_16","alias_value":"V7XL7TQMKV4RUZMM","created_at":"2026-07-05T11:06:13.252000+00:00"},{"alias_kind":"pith_short_8","alias_value":"V7XL7TQM","created_at":"2026-07-05T11:06:13.252000+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.30646","citing_title":"Same Patient, Different Words, Different Diagnosis? Evaluating Semantic Stability in Clinical LLMs","ref_index":31,"is_internal_anchor":true},{"citing_arxiv_id":"2606.18103","citing_title":"HistoRAG: Embedding Historical Methodology in Retrieval-Augmented Generation Through Critical Technical Practice","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2604.10116","citing_title":"A Dual Cross-Attention Graph Learning Framework For Multimodal MRI-Based Major Depressive Disorder Detection","ref_index":36,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN","json":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN.json","graph_json":"https://pith.science/api/pith-number/V7XL7TQMKV4RUZMMBSCWNLMFFN/graph.json","events_json":"https://pith.science/api/pith-number/V7XL7TQMKV4RUZMMBSCWNLMFFN/events.json","paper":"https://pith.science/paper/V7XL7TQM"},"agent_actions":{"view_html":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN","download_json":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN.json","view_paper":"https://pith.science/paper/V7XL7TQM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.16318&json=true","fetch_graph":"https://pith.science/api/pith-number/V7XL7TQMKV4RUZMMBSCWNLMFFN/graph.json","fetch_events":"https://pith.science/api/pith-number/V7XL7TQMKV4RUZMMBSCWNLMFFN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN/action/storage_attestation","attest_author":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN/action/author_attestation","sign_citation":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN/action/citation_signature","submit_replication":"https://pith.science/pith/V7XL7TQMKV4RUZMMBSCWNLMFFN/action/replication_record"}},"created_at":"2026-07-05T11:06:13.252000+00:00","updated_at":"2026-07-05T11:06:13.252000+00:00"}