Why A.I. Safety Controls Are Not Very Effective

kept by eddie jun 10

Jailbreaking AI systems to bypass safety controls remains trivially easy three years after ChatGPT launched, despite industry claims of progress.

Opinion | Anthropic’s Chief on A.I.: ‘We Don’t Know if the Models Are Conscious’ - The New York Times
Anthropic CEO Dario Amodei discusses uncertain AI consciousness and shares both utopian and dystopian near-term predictions for artificial intelligence.
also on #artificial-intelligence, #ai-safety
The Promise of Polymath LLMs - by Robin Hanson
Robin Hanson argues that large language models could drive progress by systematically finding and resolving contradictions between unrelated fields of knowledge.
also on #artificial-intelligence
Redeploying Claude Fable 5 \ Anthropic
Anthropic redeploys Claude Fable 5 after export controls lift, with updated safeguards and a proposed industry jailbreak severity framework.
also on #cybersecurity, #ai-safety
The Hacker Sent by Anthropic to Calm the Government’s Nerves About AI Safety - WSJ
Anthropic's Nicholas Carlini, who demonstrated AI's ability to find critical security bugs, is now arguing for releasing models to calm U.S. government concerns.
also on #cybersecurity, #ai-safety