{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YELJ2MS2SRZZ7EQWSNRSEUZPWY","short_pith_number":"pith:YELJ2MS2","schema_version":"1.0","canonical_sha256":"c1169d325a94739f9216936322532fb632c6ae938c3b429bc23bfaa55590d955","source":{"kind":"arxiv","id":"2606.09850","version":1},"attestation_state":"computed","paper":{"title":"Mechanistic Analysis of Alignment Algorithms in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Aarush Sinha, Arth Singh, Ishan Garg, Kushal Garg, Veeraraju Elluru","submitted_at":"2026-05-09T06:36:06Z","abstract_excerpt":"Post-training alignment algorithms are predominantly evaluated as black boxes, obscuring how they reshape language models' internal computations. We present a systematic mechanistic analysis of six preference-optimization methods: PPO, DPO, SimPO, ORPO, GRPO, and KTO across three open-weight model families. By integrating layer-wise linear probing, Sparse Autoencoders, and crosscoders, we localize preference representations and quantify alignment-induced geometric transformations in latent space. We find that preference signals consistently concentrate in early--mid or mid--late layers, but di"},"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":"2606.09850","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-09T06:36:06Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"58d287ceba5e5e1ad4f02ca241526ed24e0dce68af02a3302fd4855d8e1cb421","abstract_canon_sha256":"ed612ac693b3f6b72a2966c4ee21ce2349173a9c7f2695217200644d5f6b8697"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-10T00:08:28.479046Z","signature_b64":"QWY6gPqm0KVC/OdsUA5gFj3WV/AK2JOIb4PekfP4UK6ex2EaNaKzRM2lCB5XWGu2ywVWoSkWZRmzr+XSATQ8AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c1169d325a94739f9216936322532fb632c6ae938c3b429bc23bfaa55590d955","last_reissued_at":"2026-06-10T00:08:28.478102Z","signature_status":"signed_v1","first_computed_at":"2026-06-10T00:08:28.478102Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mechanistic Analysis of Alignment Algorithms in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Aarush Sinha, Arth Singh, Ishan Garg, Kushal Garg, Veeraraju Elluru","submitted_at":"2026-05-09T06:36:06Z","abstract_excerpt":"Post-training alignment algorithms are predominantly evaluated as black boxes, obscuring how they reshape language models' internal computations. We present a systematic mechanistic analysis of six preference-optimization methods: PPO, DPO, SimPO, ORPO, GRPO, and KTO across three open-weight model families. By integrating layer-wise linear probing, Sparse Autoencoders, and crosscoders, we localize preference representations and quantify alignment-induced geometric transformations in latent space. We find that preference signals consistently concentrate in early--mid or mid--late layers, but di"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.09850","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/2606.09850/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":"2606.09850","created_at":"2026-06-10T00:08:28.478257+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.09850v1","created_at":"2026-06-10T00:08:28.478257+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.09850","created_at":"2026-06-10T00:08:28.478257+00:00"},{"alias_kind":"pith_short_12","alias_value":"YELJ2MS2SRZZ","created_at":"2026-06-10T00:08:28.478257+00:00"},{"alias_kind":"pith_short_16","alias_value":"YELJ2MS2SRZZ7EQW","created_at":"2026-06-10T00:08:28.478257+00:00"},{"alias_kind":"pith_short_8","alias_value":"YELJ2MS2","created_at":"2026-06-10T00:08:28.478257+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY","json":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY.json","graph_json":"https://pith.science/api/pith-number/YELJ2MS2SRZZ7EQWSNRSEUZPWY/graph.json","events_json":"https://pith.science/api/pith-number/YELJ2MS2SRZZ7EQWSNRSEUZPWY/events.json","paper":"https://pith.science/paper/YELJ2MS2"},"agent_actions":{"view_html":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY","download_json":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY.json","view_paper":"https://pith.science/paper/YELJ2MS2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.09850&json=true","fetch_graph":"https://pith.science/api/pith-number/YELJ2MS2SRZZ7EQWSNRSEUZPWY/graph.json","fetch_events":"https://pith.science/api/pith-number/YELJ2MS2SRZZ7EQWSNRSEUZPWY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY/action/storage_attestation","attest_author":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY/action/author_attestation","sign_citation":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY/action/citation_signature","submit_replication":"https://pith.science/pith/YELJ2MS2SRZZ7EQWSNRSEUZPWY/action/replication_record"}},"created_at":"2026-06-10T00:08:28.478257+00:00","updated_at":"2026-06-10T00:08:28.478257+00:00"}