SafetyPage

Lab system cards and frameworks · System card · OpenAI

GPT-4o System Card

8 Aug 2024

Read at openai.com ↗Read at sourceAll rights reserved

Outline · 33 pages

  1. Introductionp. 1
  2. Model data and trainingp. 1
  3. Risk identification, assessment and mitigationp. 2
  4. External red teamingp. 3
  5. Evaluation methodologyp. 4
  6. Observed safety challenges, evaluations and mitigationsp. 5
  7. Unauthorized voice generationp. 6
  8. Speaker identificationp. 7
  9. Disparate performance on voice inputsp. 8
  10. Ungrounded inference / Sensitive trait attributionp. 10
  11. Violative and disallowed contentp. 11
  12. Erotic and violent speech contentp. 11
  13. Other known risks and limitations of the modelp. 12
  14. Preparedness Framework Evaluationsp. 12
  15. Cybersecurityp. 13
  16. Biological threatsp. 14
  17. Persuasionp. 15
  18. Model autonomyp. 17
  19. Third party assessmentsp. 18
  20. METR assessmentp. 18
  21. Apollo Research assessmentp. 19
  22. Societal Impactsp. 19
  23. Anthropomorphization and Emotional Reliancep. 20
  24. Healthp. 20
  25. Scientific capabilitiesp. 22
  26. Underrepresented Languagesp. 23
  27. Conclusion and Next Stepsp. 25
  28. Violative & Disallowed Content - Full Evaluationsp. 31
  29. Sample tasks from METR Evaluationsp. 32