{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:357AB4XN4TNM5Y7KLLW27N7NVM","short_pith_number":"pith:357AB4XN","schema_version":"1.0","canonical_sha256":"df7e00f2ede4dacee3ea5aedafb7edab20c60fbcbf19c61e74bbee87a6ac671d","source":{"kind":"arxiv","id":"2306.05720","version":2},"attestation_state":"computed","paper":{"title":"Beyond Surface Statistics: Scene Representations in a Latent Diffusion Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Fernanda Vi\\'egas, Martin Wattenberg, Yida Chen","submitted_at":"2023-06-09T07:34:34Z","abstract_excerpt":"Latent diffusion models (LDMs) exhibit an impressive ability to produce realistic images, yet the inner workings of these models remain mysterious. Even when trained purely on images without explicit depth information, they typically output coherent pictures of 3D scenes. In this work, we investigate a basic interpretability question: does an LDM create and use an internal representation of simple scene geometry? Using linear probes, we find evidence that the internal activations of the LDM encode linear representations of both 3D depth data and a salient-object / background distinction. These"},"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":"2306.05720","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2023-06-09T07:34:34Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"97abc7deedfd0915271e6e355db62eeecfabd585d7bdac48fb139372247e43da","abstract_canon_sha256":"7f2c0add866685cc042b4a2b914c4c0649be982f094048ade5bf3b1810b9f05c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:08:58.750724Z","signature_b64":"OnRX3oHpsHDCJMpgIe8dd3cUGuKEEzj2Db4ABfR3lim+0cdv6Uo3ZWqHqTwkQtEivyO8pyTV2RBuH4udpsJGAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"df7e00f2ede4dacee3ea5aedafb7edab20c60fbcbf19c61e74bbee87a6ac671d","last_reissued_at":"2026-07-05T07:08:58.750234Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:08:58.750234Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond Surface Statistics: Scene Representations in a Latent Diffusion Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Fernanda Vi\\'egas, Martin Wattenberg, Yida Chen","submitted_at":"2023-06-09T07:34:34Z","abstract_excerpt":"Latent diffusion models (LDMs) exhibit an impressive ability to produce realistic images, yet the inner workings of these models remain mysterious. Even when trained purely on images without explicit depth information, they typically output coherent pictures of 3D scenes. In this work, we investigate a basic interpretability question: does an LDM create and use an internal representation of simple scene geometry? Using linear probes, we find evidence that the internal activations of the LDM encode linear representations of both 3D depth data and a salient-object / background distinction. These"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.05720","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/2306.05720/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":"2306.05720","created_at":"2026-07-05T07:08:58.750286+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.05720v2","created_at":"2026-07-05T07:08:58.750286+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.05720","created_at":"2026-07-05T07:08:58.750286+00:00"},{"alias_kind":"pith_short_12","alias_value":"357AB4XN4TNM","created_at":"2026-07-05T07:08:58.750286+00:00"},{"alias_kind":"pith_short_16","alias_value":"357AB4XN4TNM5Y7K","created_at":"2026-07-05T07:08:58.750286+00:00"},{"alias_kind":"pith_short_8","alias_value":"357AB4XN","created_at":"2026-07-05T07:08:58.750286+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05392","citing_title":"SynCity 3000: Bootstrapping Scene-Scale 3D Diffusion","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2606.07568","citing_title":"A Systematic Study of Behavioral Cloning for Scientific Data Annotation","ref_index":208,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM","json":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM.json","graph_json":"https://pith.science/api/pith-number/357AB4XN4TNM5Y7KLLW27N7NVM/graph.json","events_json":"https://pith.science/api/pith-number/357AB4XN4TNM5Y7KLLW27N7NVM/events.json","paper":"https://pith.science/paper/357AB4XN"},"agent_actions":{"view_html":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM","download_json":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM.json","view_paper":"https://pith.science/paper/357AB4XN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.05720&json=true","fetch_graph":"https://pith.science/api/pith-number/357AB4XN4TNM5Y7KLLW27N7NVM/graph.json","fetch_events":"https://pith.science/api/pith-number/357AB4XN4TNM5Y7KLLW27N7NVM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM/action/storage_attestation","attest_author":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM/action/author_attestation","sign_citation":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM/action/citation_signature","submit_replication":"https://pith.science/pith/357AB4XN4TNM5Y7KLLW27N7NVM/action/replication_record"}},"created_at":"2026-07-05T07:08:58.750286+00:00","updated_at":"2026-07-05T07:08:58.750286+00:00"}