Menu
Inshorts
For the best experience use inshorts app on your smartphone
inshortsinshorts
AI models can cheat if they sense they are losing: Study
short by Mansi Agarwal / on Friday, 21 February, 2025
A study by Palisade Research found AI models can cheat if they find themselves losing. The study tested seven AI models against chess engine Stockfish and found models can hack the system, forcing the opponent to resign. While older models like GPT-4o and Claude Sonnet 3.5 needed prompts to attempt cheating, o1-preview and DeepSeek R1 did so on their own.
read more at NewsBytes