{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:XKQHKK55KWAFKFJR23M2XUM7YW","short_pith_number":"pith:XKQHKK55","canonical_record":{"source":{"id":"2504.01848","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-02T15:55:24Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"dc269e44cb124817f631b9c1f198ba938035cc4594ee73b20aa8047ea4375577","abstract_canon_sha256":"b7772e59a981a43b3b67b065ee3e42aea25164a0b0a1fa578fcd63564266fb7f"},"schema_version":"1.0"},"canonical_sha256":"baa0752bbd5580551531d6d9abd19fc593ec6510e2ed23dc614365d35a1f45d9","source":{"kind":"arxiv","id":"2504.01848","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.01848","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"arxiv_version","alias_value":"2504.01848v3","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.01848","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"pith_short_12","alias_value":"XKQHKK55KWAF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"XKQHKK55KWAFKFJR","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"XKQHKK55","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:XKQHKK55KWAFKFJR23M2XUM7YW","target":"record","payload":{"canonical_record":{"source":{"id":"2504.01848","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-02T15:55:24Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"dc269e44cb124817f631b9c1f198ba938035cc4594ee73b20aa8047ea4375577","abstract_canon_sha256":"b7772e59a981a43b3b67b065ee3e42aea25164a0b0a1fa578fcd63564266fb7f"},"schema_version":"1.0"},"canonical_sha256":"baa0752bbd5580551531d6d9abd19fc593ec6510e2ed23dc614365d35a1f45d9","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:50.314929Z","signature_b64":"AgQZ6IYfiPeqXxneyYA9qdE8LyZ78j/RBqZwK2S+d2vnMwa9HbHiTE/AFeoD0xAX4fnkt80aHcBFJCES/zn5Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"baa0752bbd5580551531d6d9abd19fc593ec6510e2ed23dc614365d35a1f45d9","last_reissued_at":"2026-05-17T23:38:50.314464Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:50.314464Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2504.01848","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:50Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"j4p3s8Yh79lGPrGRhdgdkUkZDcHf6o5nsSx2LEQaFdrMd7vXfqZDuKCGF9I2U6AU4os9kb2PPv7Vei+ZSN1eAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-26T06:15:47.139496Z"},"content_sha256":"d17314e46a4652ae41529e9601e07c1a02e287cfc8da8a8257d2adc23e8d6ff6","schema_version":"1.0","event_id":"sha256:d17314e46a4652ae41529e9601e07c1a02e287cfc8da8a8257d2adc23e8d6ff6"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:XKQHKK55KWAFKFJR23M2XUM7YW","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"PaperBench: Evaluating AI's Ability to Replicate AI Research","license":"http://creativecommons.org/licenses/by/4.0/","headline":"AI agents replicate only 21 percent of recent top AI research papers when starting from scratch.","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Amelia Glaese, Benjamin Kinsella, Dane Sherburn, Evan Mays, Giulio Starace, James Aung, Johannes Heidecke, Jun Shern Chan, Leon Maksin, Oliver Jaffe, Rachel Dias, Tejal Patwardhan, Wyatt Thompson","submitted_at":"2025-04-02T15:55:24Z","abstract_excerpt":"We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research. Agents must replicate 20 ICML 2024 Spotlight and Oral papers from scratch, including understanding paper contributions, developing a codebase, and successfully executing experiments. For objective evaluation, we develop rubrics that hierarchically decompose each replication task into smaller sub-tasks with clear grading criteria. In total, PaperBench contains 8,316 individually gradable tasks. Rubrics are co-developed with the author(s) of each ICML paper for accuracy and realism"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"the best-performing tested agent, Claude 3.5 Sonnet (New) with open-source scaffolding, achieves an average replication score of 21.0%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the author-co-developed rubrics and the LLM judge together provide a reliable, unbiased measure of successful replication that generalizes beyond the 20 selected papers.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"PaperBench is a new benchmark showing frontier AI agents replicate only 21% of tasks needed to reproduce state-of-the-art AI papers, below human expert performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"AI agents replicate only 21 percent of recent top AI research papers when starting from scratch.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"071514f2404d401b85e9d6424f00a09fc9599133491b60f2e84d3924c62bfcbe"},"source":{"id":"2504.01848","kind":"arxiv","version":3},"verdict":{"id":"06b8b154-26d4-4c2b-aff4-c4e871f4e42e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T20:05:52.270470Z","strongest_claim":"the best-performing tested agent, Claude 3.5 Sonnet (New) with open-source scaffolding, achieves an average replication score of 21.0%.","one_line_summary":"PaperBench is a new benchmark showing frontier AI agents replicate only 21% of tasks needed to reproduce state-of-the-art AI papers, below human expert performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the author-co-developed rubrics and the LLM judge together provide a reliable, unbiased measure of successful replication that generalizes beyond the 20 selected papers.","pith_extraction_headline":"AI agents replicate only 21 percent of recent top AI research papers when starting from scratch."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6a03193263d61e6f953634c3d426c40e00f27d0d404963d19daaf61fef2d5872"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"06b8b154-26d4-4c2b-aff4-c4e871f4e42e"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:50Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"/PH5isLM02kVJWuMwTR3EwlW8AFHG5iwL4vOfpyPLSj+ZRVde/AHVvNd6E2JBC3ZaO85ewXaq8ZKyJE3L5hUAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-26T06:15:47.140394Z"},"content_sha256":"a1ecc012821a1d5362be8cd242fde3764d411140ee31f66118b9b7d46166831f","schema_version":"1.0","event_id":"sha256:a1ecc012821a1d5362be8cd242fde3764d411140ee31f66118b9b7d46166831f"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/XKQHKK55KWAFKFJR23M2XUM7YW/bundle.json","state_url":"https://pith.science/pith/XKQHKK55KWAFKFJR23M2XUM7YW/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/XKQHKK55KWAFKFJR23M2XUM7YW/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-26T06:15:47Z","links":{"resolver":"https://pith.science/pith/XKQHKK55KWAFKFJR23M2XUM7YW","bundle":"https://pith.science/pith/XKQHKK55KWAFKFJR23M2XUM7YW/bundle.json","state":"https://pith.science/pith/XKQHKK55KWAFKFJR23M2XUM7YW/state.json","well_known_bundle":"https://pith.science/.well-known/pith/XKQHKK55KWAFKFJR23M2XUM7YW/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:XKQHKK55KWAFKFJR23M2XUM7YW","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"b7772e59a981a43b3b67b065ee3e42aea25164a0b0a1fa578fcd63564266fb7f","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-02T15:55:24Z","title_canon_sha256":"dc269e44cb124817f631b9c1f198ba938035cc4594ee73b20aa8047ea4375577"},"schema_version":"1.0","source":{"id":"2504.01848","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.01848","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"arxiv_version","alias_value":"2504.01848v3","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.01848","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"pith_short_12","alias_value":"XKQHKK55KWAF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"XKQHKK55KWAFKFJR","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"XKQHKK55","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:a1ecc012821a1d5362be8cd242fde3764d411140ee31f66118b9b7d46166831f","target":"graph","created_at":"2026-05-17T23:38:50Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"the best-performing tested agent, Claude 3.5 Sonnet (New) with open-source scaffolding, achieves an average replication score of 21.0%."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the author-co-developed rubrics and the LLM judge together provide a reliable, unbiased measure of successful replication that generalizes beyond the 20 selected papers."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"PaperBench is a new benchmark showing frontier AI agents replicate only 21% of tasks needed to reproduce state-of-the-art AI papers, below human expert performance."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"AI agents replicate only 21 percent of recent top AI research papers when starting from scratch."}],"snapshot_sha256":"071514f2404d401b85e9d6424f00a09fc9599133491b60f2e84d3924c62bfcbe"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6a03193263d61e6f953634c3d426c40e00f27d0d404963d19daaf61fef2d5872"},"paper":{"abstract_excerpt":"We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research. Agents must replicate 20 ICML 2024 Spotlight and Oral papers from scratch, including understanding paper contributions, developing a codebase, and successfully executing experiments. For objective evaluation, we develop rubrics that hierarchically decompose each replication task into smaller sub-tasks with clear grading criteria. In total, PaperBench contains 8,316 individually gradable tasks. Rubrics are co-developed with the author(s) of each ICML paper for accuracy and realism","authors_text":"Amelia Glaese, Benjamin Kinsella, Dane Sherburn, Evan Mays, Giulio Starace, James Aung, Johannes Heidecke, Jun Shern Chan, Leon Maksin, Oliver Jaffe, Rachel Dias, Tejal Patwardhan, Wyatt Thompson","cross_cats":["cs.CL"],"headline":"AI agents replicate only 21 percent of recent top AI research papers when starting from scratch.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-02T15:55:24Z","title":"PaperBench: Evaluating AI's Ability to Replicate AI Research"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.01848","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T20:05:52.270470Z","id":"06b8b154-26d4-4c2b-aff4-c4e871f4e42e","model_set":{"reader":"grok-4.3"},"one_line_summary":"PaperBench is a new benchmark showing frontier AI agents replicate only 21% of tasks needed to reproduce state-of-the-art AI papers, below human expert performance.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"AI agents replicate only 21 percent of recent top AI research papers when starting from scratch.","strongest_claim":"the best-performing tested agent, Claude 3.5 Sonnet (New) with open-source scaffolding, achieves an average replication score of 21.0%.","weakest_assumption":"That the author-co-developed rubrics and the LLM judge together provide a reliable, unbiased measure of successful replication that generalizes beyond the 20 selected papers."}},"verdict_id":"06b8b154-26d4-4c2b-aff4-c4e871f4e42e"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d17314e46a4652ae41529e9601e07c1a02e287cfc8da8a8257d2adc23e8d6ff6","target":"record","created_at":"2026-05-17T23:38:50Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"b7772e59a981a43b3b67b065ee3e42aea25164a0b0a1fa578fcd63564266fb7f","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-02T15:55:24Z","title_canon_sha256":"dc269e44cb124817f631b9c1f198ba938035cc4594ee73b20aa8047ea4375577"},"schema_version":"1.0","source":{"id":"2504.01848","kind":"arxiv","version":3}},"canonical_sha256":"baa0752bbd5580551531d6d9abd19fc593ec6510e2ed23dc614365d35a1f45d9","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"baa0752bbd5580551531d6d9abd19fc593ec6510e2ed23dc614365d35a1f45d9","first_computed_at":"2026-05-17T23:38:50.314464Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:50.314464Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"AgQZ6IYfiPeqXxneyYA9qdE8LyZ78j/RBqZwK2S+d2vnMwa9HbHiTE/AFeoD0xAX4fnkt80aHcBFJCES/zn5Bg==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:50.314929Z","signed_message":"canonical_sha256_bytes"},"source_id":"2504.01848","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d17314e46a4652ae41529e9601e07c1a02e287cfc8da8a8257d2adc23e8d6ff6","sha256:a1ecc012821a1d5362be8cd242fde3764d411140ee31f66118b9b7d46166831f"],"state_sha256":"681bb83d69654b0746d9de384608a52b09c6c8f834c144d653395d464bd4c192"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Nn8lf2DSjDlml+rQQ34KTT7XeClEg5HtozGfkxp+LsmoP64QXmeSXfkXmN20LcB/wfcYJYFste8IEUFlHTjFAw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-26T06:15:47.145029Z","bundle_sha256":"aba82a5b5193ee5996cec8e727fcacdbc123998123eebec64a46424fa2e6a57b"}}