{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:OPUXJ6LJA62WSPGL5LZXJVZ4AJ","short_pith_number":"pith:OPUXJ6LJ","schema_version":"1.0","canonical_sha256":"73e974f96907b5693ccbeaf374d73c027ac7ae8776fcc4d964193e9f6428c634","source":{"kind":"arxiv","id":"2003.12298","version":1},"attestation_state":"computed","paper":{"title":"Information-Theoretic Probing with Minimum Description Length","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Elena Voita, Ivan Titov","submitted_at":"2020-03-27T09:35:38Z","abstract_excerpt":"To measure how well pretrained representations encode some linguistic property, it is common to use accuracy of a probe, i.e. a classifier trained to predict the property from the representations. Despite widespread adoption of probes, differences in their accuracy fail to adequately reflect differences in representations. For example, they do not substantially favour pretrained representations over randomly initialized ones. Analogously, their accuracy can be similar when probing for genuine linguistic labels and probing for random synthetic tasks. To see reasonable differences in accuracy wi"},"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":"2003.12298","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-03-27T09:35:38Z","cross_cats_sorted":[],"title_canon_sha256":"e962e6e1117bf609cc04359895a4c2c03c47f76dca6cf59aacafd25e5c830155","abstract_canon_sha256":"b13b1032286302ad4de9069ee993fd33f0c0101e295699ba33c7daaa85cf7271"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:50:56.972726Z","signature_b64":"JoNrdjLYiu+475pZEWWnNdlOFn+YTFnbotFDD/zSk7h7tRweZYuSuUpIHxWEPes+8bDhoMYzBBcXq3aEj7ooAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"73e974f96907b5693ccbeaf374d73c027ac7ae8776fcc4d964193e9f6428c634","last_reissued_at":"2026-07-05T00:50:56.972243Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:50:56.972243Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Information-Theoretic Probing with Minimum Description Length","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Elena Voita, Ivan Titov","submitted_at":"2020-03-27T09:35:38Z","abstract_excerpt":"To measure how well pretrained representations encode some linguistic property, it is common to use accuracy of a probe, i.e. a classifier trained to predict the property from the representations. Despite widespread adoption of probes, differences in their accuracy fail to adequately reflect differences in representations. For example, they do not substantially favour pretrained representations over randomly initialized ones. Analogously, their accuracy can be similar when probing for genuine linguistic labels and probing for random synthetic tasks. To see reasonable differences in accuracy wi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2003.12298","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/2003.12298/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":"2003.12298","created_at":"2026-07-05T00:50:56.972302+00:00"},{"alias_kind":"arxiv_version","alias_value":"2003.12298v1","created_at":"2026-07-05T00:50:56.972302+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2003.12298","created_at":"2026-07-05T00:50:56.972302+00:00"},{"alias_kind":"pith_short_12","alias_value":"OPUXJ6LJA62W","created_at":"2026-07-05T00:50:56.972302+00:00"},{"alias_kind":"pith_short_16","alias_value":"OPUXJ6LJA62WSPGL","created_at":"2026-07-05T00:50:56.972302+00:00"},{"alias_kind":"pith_short_8","alias_value":"OPUXJ6LJ","created_at":"2026-07-05T00:50:56.972302+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22873","citing_title":"SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning","ref_index":200,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11375","citing_title":"When Probing Accuracy Saturates, Fragility Resolves: A Complementary Metric for LLM Pre-Training Analysis","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22873","citing_title":"SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning","ref_index":199,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05741","citing_title":"HyperLens: Quantifying Cognitive Effort in LLMs with Fine-grained Confidence Trajectory","ref_index":32,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ","json":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ.json","graph_json":"https://pith.science/api/pith-number/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/graph.json","events_json":"https://pith.science/api/pith-number/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/events.json","paper":"https://pith.science/paper/OPUXJ6LJ"},"agent_actions":{"view_html":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ","download_json":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ.json","view_paper":"https://pith.science/paper/OPUXJ6LJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2003.12298&json=true","fetch_graph":"https://pith.science/api/pith-number/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/graph.json","fetch_events":"https://pith.science/api/pith-number/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/action/storage_attestation","attest_author":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/action/author_attestation","sign_citation":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/action/citation_signature","submit_replication":"https://pith.science/pith/OPUXJ6LJA62WSPGL5LZXJVZ4AJ/action/replication_record"}},"created_at":"2026-07-05T00:50:56.972302+00:00","updated_at":"2026-07-05T00:50:56.972302+00:00"}