{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:6WJKGJBCS36PSMWAFF3LO7T5S5","short_pith_number":"pith:6WJKGJBC","schema_version":"1.0","canonical_sha256":"f592a3242296fcf932c02976b77e7d974c7d73a1818a50aa1ab2317fbabc3cdf","source":{"kind":"arxiv","id":"2605.23033","version":1},"attestation_state":"computed","paper":{"title":"Uncovering the Latent Potential of Deep Intermediate Representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aniket Khandelwal, Anubha Gupta, Arnesh Batra, Arush Gumber, Jashn Khemani","submitted_at":"2026-05-21T20:58:42Z","abstract_excerpt":"Foundational Models pretrained on huge amount of data learn representations that evolve across depth, forming a hierarchy of embeddings with distinct semantic content and geometric structure. Contrary to the widespread practice of using only the final layer or shallow mixtures, we show that task-relevant information is distributed non-monotonically across layers and cannot be recovered by na\\\"ive aggregation. Through a geometric and empirical study across multiple modalities, we show that effective transfer depends on identifying which layers encode task-discriminative structure and how their "},"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":"2605.23033","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-21T20:58:42Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"a4f6ec65717acc2d53666b481b30a1359c945a2235e4fa28928344a48c3d7609","abstract_canon_sha256":"98f30cadf0bd584ccc2cb776f5bab4bb0beb223c4d7c42882ede7150c2421c3d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:35.469575Z","signature_b64":"xfGjM167CC4DRVnv9Hz9wERrl9pWp8r7fjiMRVDdrJIs4AcYmhJtN7xKBawJuH2Cm8l9T0bakeXoxt4WGML5Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f592a3242296fcf932c02976b77e7d974c7d73a1818a50aa1ab2317fbabc3cdf","last_reissued_at":"2026-05-25T02:01:35.468924Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:35.468924Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Uncovering the Latent Potential of Deep Intermediate Representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aniket Khandelwal, Anubha Gupta, Arnesh Batra, Arush Gumber, Jashn Khemani","submitted_at":"2026-05-21T20:58:42Z","abstract_excerpt":"Foundational Models pretrained on huge amount of data learn representations that evolve across depth, forming a hierarchy of embeddings with distinct semantic content and geometric structure. Contrary to the widespread practice of using only the final layer or shallow mixtures, we show that task-relevant information is distributed non-monotonically across layers and cannot be recovered by na\\\"ive aggregation. Through a geometric and empirical study across multiple modalities, we show that effective transfer depends on identifying which layers encode task-discriminative structure and how their "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.23033","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/2605.23033/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":"2605.23033","created_at":"2026-05-25T02:01:35.469022+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.23033v1","created_at":"2026-05-25T02:01:35.469022+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.23033","created_at":"2026-05-25T02:01:35.469022+00:00"},{"alias_kind":"pith_short_12","alias_value":"6WJKGJBCS36P","created_at":"2026-05-25T02:01:35.469022+00:00"},{"alias_kind":"pith_short_16","alias_value":"6WJKGJBCS36PSMWA","created_at":"2026-05-25T02:01:35.469022+00:00"},{"alias_kind":"pith_short_8","alias_value":"6WJKGJBC","created_at":"2026-05-25T02:01:35.469022+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/6WJKGJBCS36PSMWAFF3LO7T5S5","json":"https://pith.science/pith/6WJKGJBCS36PSMWAFF3LO7T5S5.json","graph_json":"https://pith.science/api/pith-number/6WJKGJBCS36PSMWAFF3LO7T5S5/graph.json","events_json":"https://pith.science/api/pith-number/6WJKGJBCS36PSMWAFF3LO7T5S5/events.json","paper":"https://pith.science/paper/6WJKGJBC"},"agent_actions":{"view_html":"https://pith.science/pith/6WJKGJBCS36PSMWAFF3LO7T5S5","download_json":"https://pith.science/pith/6WJKGJBCS36PSMWAFF3LO7T5S5.json","view_paper":"https://pith.science/paper/6WJKGJBC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.23033&json=true","fetch_graph":"https://pith.science/api/pith-number/6WJKGJBCS36PSMWAFF3LO7T5S5/graph.json","fetch_events":"https://pith.science/api/pith-number/6WJKGJBCS36PSMWAFF3LO7T5S5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6WJKGJBCS36PSMWAFF3LO7T5S5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6WJKGJBCS36PSMWAFF3LO7T5S5/action/storage_attestation","attest_author":"https://pith.science/pith/6WJKGJBCS36PSMWAFF3LO7T5S5/action/author_attestation","sign_citation":"https://pith.science/pith/6WJKGJBCS36PSMWAFF3LO7T5S5/action/citation_signature","submit_replication":"https://pith.science/pith/6WJKGJBCS36PSMWAFF3LO7T5S5/action/replication_record"}},"created_at":"2026-05-25T02:01:35.469022+00:00","updated_at":"2026-05-25T02:01:35.469022+00:00"}