{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:4O3AV5JBILMAJ7AJ7JO6FBKR3F","short_pith_number":"pith:4O3AV5JB","schema_version":"1.0","canonical_sha256":"e3b60af52142d804fc09fa5de28551d96b87722bd7ec9d9050d6a983c507aa58","source":{"kind":"arxiv","id":"2502.12150","version":2},"attestation_state":"computed","paper":{"title":"Idiosyncrasies in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"J. Zico Kolter, Mingjie Sun, Yida Yin, Zhiqiu Xu, Zhuang Liu","submitted_at":"2025-02-17T18:59:02Z","abstract_excerpt":"In this work, we unveil and study idiosyncrasies in Large Language Models (LLMs) -- unique patterns in their outputs that can be used to distinguish the models. To do so, we consider a simple classification task: given a particular text output, the objective is to predict the source LLM that generates the text. We evaluate this synthetic task across various groups of LLMs and find that simply fine-tuning text embedding models on LLM-generated texts yields excellent classification accuracy. Notably, we achieve 97.1% accuracy on held-out validation data in the five-way classification problem inv"},"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":"2502.12150","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-02-17T18:59:02Z","cross_cats_sorted":[],"title_canon_sha256":"56840904bdf6f1584f02c6a24df1e722e5a08fecc1f4639e774eef2ebe9b985c","abstract_canon_sha256":"676e30e95e55e6a59184025e02d0395250475b0dcaf8838ea60c549a3dde876d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:22:17.780721Z","signature_b64":"Ddz4ZmP6VE6Uc3vEbUmPd4qUb4fY4nZAc3E5rS7mxVAmf+jB6h1qeT9dUG58LUGanw6hpwt7fNT3ZV1mLeWSBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e3b60af52142d804fc09fa5de28551d96b87722bd7ec9d9050d6a983c507aa58","last_reissued_at":"2026-07-05T11:22:17.780210Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:22:17.780210Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Idiosyncrasies in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"J. Zico Kolter, Mingjie Sun, Yida Yin, Zhiqiu Xu, Zhuang Liu","submitted_at":"2025-02-17T18:59:02Z","abstract_excerpt":"In this work, we unveil and study idiosyncrasies in Large Language Models (LLMs) -- unique patterns in their outputs that can be used to distinguish the models. To do so, we consider a simple classification task: given a particular text output, the objective is to predict the source LLM that generates the text. We evaluate this synthetic task across various groups of LLMs and find that simply fine-tuning text embedding models on LLM-generated texts yields excellent classification accuracy. Notably, we achieve 97.1% accuracy on held-out validation data in the five-way classification problem inv"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12150","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/2502.12150/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":"2502.12150","created_at":"2026-07-05T11:22:17.780269+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.12150v2","created_at":"2026-07-05T11:22:17.780269+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12150","created_at":"2026-07-05T11:22:17.780269+00:00"},{"alias_kind":"pith_short_12","alias_value":"4O3AV5JBILMA","created_at":"2026-07-05T11:22:17.780269+00:00"},{"alias_kind":"pith_short_16","alias_value":"4O3AV5JBILMAJ7AJ","created_at":"2026-07-05T11:22:17.780269+00:00"},{"alias_kind":"pith_short_8","alias_value":"4O3AV5JB","created_at":"2026-07-05T11:22:17.780269+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.16462","citing_title":"Asking Back: Interaction-Layer Antidistillation Watermarks","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2509.22055","citing_title":"RedNote-Vibe: A Dataset for Capturing Temporal Dynamics of AI-Generated Text in Lifestyle Social Media","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08276","citing_title":"ACF: A Collaborative Framework for Agent Covert Communication under Cognitive Asymmetry","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16755","citing_title":"Machine individuality: Separating genuine idiosyncrasy from response bias in large language models","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F","json":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F.json","graph_json":"https://pith.science/api/pith-number/4O3AV5JBILMAJ7AJ7JO6FBKR3F/graph.json","events_json":"https://pith.science/api/pith-number/4O3AV5JBILMAJ7AJ7JO6FBKR3F/events.json","paper":"https://pith.science/paper/4O3AV5JB"},"agent_actions":{"view_html":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F","download_json":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F.json","view_paper":"https://pith.science/paper/4O3AV5JB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.12150&json=true","fetch_graph":"https://pith.science/api/pith-number/4O3AV5JBILMAJ7AJ7JO6FBKR3F/graph.json","fetch_events":"https://pith.science/api/pith-number/4O3AV5JBILMAJ7AJ7JO6FBKR3F/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F/action/storage_attestation","attest_author":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F/action/author_attestation","sign_citation":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F/action/citation_signature","submit_replication":"https://pith.science/pith/4O3AV5JBILMAJ7AJ7JO6FBKR3F/action/replication_record"}},"created_at":"2026-07-05T11:22:17.780269+00:00","updated_at":"2026-07-05T11:22:17.780269+00:00"}