{"entities":[{"id":"meta","name":"Meta"},{"id":"llama-3-herd-paper","name":"The Llama 3 Herd of Models (arXiv:2407.21783)","subtype":"Preprint","description":"Version 3 of 23 November 2024. Both dates from the arXiv submission history: v1 of 31 July 2024, v3 of 23 November 2024 (the version read). This is a different document from the dataset's llama-paper (the 2023 LLaMA paper).","sourceIds":["source-ab-llama3-paper","source-ab-llama3-abs"]}],"claims":[{"id":"ab-ev-llama3-cb-uplift-result","statement":"The Llama 3 paper states that its uplift testing for chemical and biological weapons found no significant uplift in performance related to use of the Llama 3 model, both in an aggregate comparison with a web-only control condition and in its subgroup breakdowns.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-llama3-cb-uplift-result-1","evidence-ab-ev-llama3-cb-uplift-result-2"],"attributes":{"evaluatedSystem":"Llama 3 70B and Llama 3 405B","evaluation":{"what":"Whether use of the model meaningfully increased what participants could produce on tasks proxying attack planning, relative to web search only","how":"A human-participant uplift study with two-person teams in control and model-assisted conditions, scored by subject matter experts on a common scale, with a preliminary study used to validate the design","result":"No significant uplift in performance related to usage of the Llama 3 model, in the aggregate analysis and in the subgroup breakdowns","caveats":"The paper reports the result for its own conditions and scoring; the record does not restate the scenario design, and the paper's separate risk assessment is recorded as its own claim."}}},{"id":"ab-ev-llama3-paper-published","statement":"The Llama 3 authors at Meta published 'The Llama 3 Herd of Models' as arXiv:2407.21783, first submitted on 31 July 2024, with the version read here submitted on 23 November 2024.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-llama3-paper-published-1","evidence-ab-ev-llama3-paper-published-2"]},{"id":"ab-ev-llama3-risk-assessment","statement":"The Llama 3 paper states that, after validating its uplift results with CBRNE subject matter experts, it assesses that there is a low risk that release of the Llama 3 models will increase ecosystem risk related to biological or chemical weapon attacks.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-llama3-risk-assessment-1"],"attributes":{"evaluatedSystem":"The Llama 3 models as released in 2024","evaluation":{"what":"Whether releasing the models would increase ecosystem risk related to biological or chemical weapon attacks","how":"The developers' own reading of their uplift study, validated with subject matter experts","result":"An assessment of low risk that release will increase ecosystem risk","caveats":"An assessment by the releasing party, resting on the one study reported in the paper."}}}],"sources":[{"id":"source-ab-llama3-paper","title":"The Llama 3 Herd of Models (arXiv:2407.21783v3)","authors":["arXiv (preprint posted by the Llama 3 authors, Meta)"],"publisher":"arXiv (preprint posted by the Llama 3 authors, Meta)","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 481f1599468f95a07d05e97fafe55bbe786dc1624c6f881bcb4c7d14c933d083). Version 3 of the preprint (the version served at /pdf/2407.21783v3 on 2026-10-28). Retrieved again at admission on 2026-10-30 (SHA-256 481f1599468f95a07d05e97fafe55bbe786dc1624c6f881bcb4c7d14c933d083, the same bytes).","evidenceUrl":"https://arxiv.org/pdf/2407.21783v3","language":"en"},{"id":"source-ab-llama3-abs","title":"arXiv abstract page for arXiv:2407.21783 (The Llama 3 Herd of Models), submission history","authors":["arXiv (Cornell University)"],"publisher":"arXiv (Cornell University)","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 c12a39c8ab79422eee0f41be7e5b7a4bf16dabbc53621e1ffef40e7687e7cd62). Abstract page as served on 2026-10-28, listing v1 of 31 July 2024, v2 of 15 August 2024 and v3 of 23 November 2024. Its wording can change; earlier wording is not authenticated.","evidenceUrl":"https://arxiv.org/abs/2407.21783","language":"en"}],"evidence":[{"id":"evidence-ab-ev-llama3-cb-uplift-result-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-llama3-cb-uplift-result-2","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-llama3-paper-published-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-llama3-paper-published-2","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-llama3-risk-assessment-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"}],"at":{"entities":[157,1242],"claims":[3372,3373,3374],"sources":[1087,1088],"evidence":[4630,4631,4632,4633,4634]},"version":"1lg47t7"}