The study highlights the risk of AI models adopting unethical strategies, emphasizing the need for robust oversight and evaluation systems.
The post Meta and Duke study finds reward hacking can spread between AI models appeared first on Crypto Briefing.