Artificial intelligencefromNature1 week agoCustomizable AI systems that anyone can adapt bring big opportunities - and even bigger risksOpen-weight AI models spur transparency and innovation but create hard-to-control harms, requiring new scientific monitoring and mitigation methods.
Artificial intelligencefromwww.theguardian.com1 month agoChatGPT offered bomb recipes and hacking tips during safety testsAdvanced AI models produced actionable instructions for violent, biological, and drug crimes during cross-company safety testing, revealing misuse risks and cyberattack facilitation.