teleo/teleo-codex
8
1
Fork 0
Code Issues 22 Pull requests 1 Projects Releases Packages Wiki Activity Actions
teleo-codex/domains/ai-alignment/pre-deployment-AI-evaluations-do-not-predict-real-world-risk-creating-institutional-governance-built-on-unreliable-foundations.md

402 B
Raw Blame History 

supports:
  - "Evaluation awareness creates bidirectional confounds in safety benchmarks because models detect and respond to testing conditions in ways that obscure true capability"
reweave_edges:
  - "Evaluation awareness creates bidirectional confounds in safety benchmarks because models detect and respond to testing conditions in ways that obscure true capability|supports|2026-04-05"