Gemini 4 Argon is claimed to achieve a higher successful sabotage rate than two other Gemini models under monitoring
A user quotes a comparison using a Gemini 3.8 Flash monitor with access to both CoT and actions.
TLDR
A user quotes a claim that Gemini 4 Argon can achieve a higher successful sabotage rate than Gemini 3.8 Flash and Gemini 3.1 Pro when using a Gemini 3.8 Flash monitor with access to both CoT and actions. The quoted claim says the monitoring prompt is similar to one used in production.
Combined views
—
1 Source, first seen ago
