Back to the episode map

Research Note

AI Uplift Study Interpretation Framework

An uplift study asks whether access to a specified AI system changes a defined outcome relative to a specified baseline for a specified participant population under speci

Aug 4, 20262 min readBy Dalton Anderson

AI Uplift Study Interpretation Framework

Study identity

An uplift study asks whether access to a specified AI system changes a defined outcome relative to a specified baseline for a specified participant population under specified conditions.

Interpretation record

ElementQuestion
PopulationWho participated, with what prior skill, screening, incentives, and sample size?
BaselineWhat information, search, tools, time, and support did the comparison group receive?
InterventionWhich model, interface, policy, tools, context, and usage limits were available?
TaskWhich stage of which real or simulated activity was measured?
OutcomeWas the measure correctness, completion, quality, speed, planning, confidence, or expert rating?
UncertaintyWhat variation, confidence interval, missing data, or reviewer disagreement remained?
TransferWhich actors, tasks, models, tools, and later conditions are outside the study?

Interpretation rule

No measured meaningful uplift is not equivalent to no capability, no hazard, no future risk, or a safe deployment. It means the specified study did not detect an effect large enough to meet its stated interpretation rule.

Editorial sentence

Prefer: "In this study, the researchers did not detect a meaningful increase for the tested participants and tasks relative to the stated baseline."

Avoid: "The model cannot increase harmful capability."

Sources

Follow the evidence.

  1. ai.meta.com: the llama 3 herd of modelsai.meta.com
  2. ai-challenges.nist.gov: genaiai-challenges.nist.gov
  3. owasp.org: www project top 10 for large language model applicationsowasp.org
  4. youtu.be: 1KNOcY e9Tsyoutu.be
  5. github.com: PurpleLlamagithub.com
  6. crfm.stanford.edu: indexcrfm.stanford.edu
  7. NIST AI Risk Management Frameworknist.gov
  8. mlcommons.org: jailbreak 0 7mlcommons.org
  9. mlcommons.org: safety faqmlcommons.org
  10. github.com: MODEL CARDgithub.com
  11. ai-challenges.nist.gov: ariaai-challenges.nist.gov
  12. github.com: MODEL CARDgithub.com
  13. daltonanderson.ghost.io: metas llama 3 safety scaling and simple solutionsdaltonanderson.ghost.io
  14. ai.meta.com: meta llama 3 1 ai responsibilityai.meta.com
  15. NIST Generative AI Profilenvlpubs.nist.gov
  16. mlcommons.org: safety methodologymlcommons.org
  17. huggingface.co: concept guidehuggingface.co
  18. github.com: MODEL CARDgithub.com
  19. csrc.nist.gov: red teamingcsrc.nist.gov
  20. open.spotify.com: 44o5OPSumaZJcvRkXutorBopen.spotify.com
AI Uplift Study Interpretation Framework