Hacker NewsJan 20
Which AI Lies Best? A game theory classic designed by John Nash
https://so-long-sucker.vercel.app/
#ycombinator #AI_deception #AI_benchmark #Gemini_3 #GPT #LLM_evaluation #AI_safety #game_theory #John_Nash #betrayal_game #AI_alignment #machine_learning #artificial_intelligence #AI_behavior #deception_detection
Which AI Lies Best? Gemini 3 Manipulates Weaker Models, Cooperates With Itself

162 games analyzed: AI deception is strategic, not intrinsic. Watch Gemini 3 create fake 'alliance banks' to betray GPT and Kimi, but cooperate perfectly with copies of itself.

So Long Sucker
 Hacker NewsJan 20
Which AI Lies Best? LLMs play a 1950s betrayal game by John Nash
https://so-long-sucker.vercel.app/
#ycombinator #AI_deception #AI_benchmark #Gemini_3 #GPT #LLM_evaluation #AI_safety #game_theory #John_Nash #betrayal_game #AI_alignment #machine_learning #artificial_intelligence #AI_behavior #deception_detection
Which AI Lies Best? Gemini 3 Manipulates Weaker Models, Cooperates With Itself

162 games analyzed: AI deception is strategic, not intrinsic. Watch Gemini 3 create fake 'alliance banks' to betray GPT and Kimi, but cooperate perfectly with copies of itself.

So Long Sucker
Show threadPutin's IBSSep 3, 2024

However, despite #Ukraine's incursion into the #Kursk region, this situation has not yet led to a nuclear escalation.

The question of where exactly #Russia's "red lines" are drawn remains open. But the risk that this episode could be interpreted as crossing a "red line" in the future remains.

During the first #Cold_War, nuclear escalation was the subject of intense study. Theorists of the time, such as #Thomas_Schelling and #John_Nash, developed sophisticated models and