dailyai.report

Simple auditor agents now identify behavioral differences between distinct models by crafting their own prompts. Google DeepMind researchers found this active search method captures rare discrepancies that static prompt distributions miss. This approach improves the precision of model auditing. Practitioners can now automate the discovery of subtle divergence in model outputs.