specific method · filed under evaluation
Dangerous-capability uplift assessment
Assesses knowledge and procedural uplift in CBRN and cyber capabilities through internal evaluations, external testing, and refusal-suppressed variants.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
For CBRN and cyber, we assessed knowledge and procedural uplift through internal evaluations, external testing, and refusal-suppressed variants
usedevaluation onlyin InklingThinking Machines Lab
Filed alongside
Other methods under evaluation :: human & real-world evaluation.
Blind side-by-side expert evaluationMulti-turn open-ended external red-teamingInternal engineering task evaluationA/B testing on real tasksAutomated red-teaming loopContinuous monitoringCyber range exercisesExternal expert review of safety evaluation methodologyHeld-out evaluationHuman evaluationPairwise comparison for Chinese writingPre-release safety evaluationReal-world codebase testingReal-world feedback refinementReward hacking mitigation in evaluationsSame-prompt comparative DevOps testingSystematic robustness-boundary characterizationWhite-collar enterprise task evaluation