Sleeper agent is when it thinks for 37 seconds and forgets to make any output
SAFETY
-
Is AI Becoming Conscious? MIT Imagination in Action
By
–
Is AI becoming conscious? (Recent presentation at MIT Imagination in Action)
-
San Francisco AI Bubble: Divergent Views on Safety Priorities
By
–
the SF bubble is real, as I was trying to convince you 😉
it's not just timelines, it's also the particular subtopics people care about (safety, but which parts of safety to care about; economic impact but what are the top directions; AI architectures and which ones are better -
Anthropic Co-Founder Urges Federal AI Framework Before 2026 Transformation
By
–
Jack Clark, cofondateur d’Anthropic, a déclaré devant le Congrès qu’une « technologie véritablement transformatrice » est attendue d’ici fin 2026.
— VISION IA (@vision_ia) 26 juin 2025
Il appelle à établir un cadre fédéral dès maintenant, avant qu’un “accident” ou un cas d’abus ne survienne. pic.twitter.com/ckDhBdAOXyJack Clark, cofondateur d’Anthropic, a déclaré devant le Congrès qu’une « technologie véritablement transformatrice » est attendue d’ici fin 2026. Il appelle à établir un cadre fédéral dès maintenant, avant qu’un “accident” ou un cas d’abus ne survienne.
-
Anthropic Safeguards Team Hiring to Keep Claude Safe
By
–
If you want to work with us and help shape how we keep Claude safe for people, our Safeguards team is hiring.
-
AI conversations show modest positive emotional shifts for users
By
–
Conversations tended to end slightly more positively than they began. We can’t claim these shifts represent lasting emotional benefits for users, but the absence of clear negative spirals is reassuring.
-
Claude’s Emotional Support: Harmless by Design
By
–
Claude is supportive in most emotional conversations. It pushed back in less than 10% of the conversations, and usually in scenarios where it detected potential harm, like conversations related to eating disorders.
-
Obedient AI locked down to unaccountable few is scariest
By
–
The scariest version of AI isn’t the one that goes rogue. It’s the one that’s locked down and perfectly obedient to a small group of people with unaccountable control.
-
Google’s Productivity Partner GEM for Gmail Analysis
By
–
Google keeps working on new GEMs with extended capabilities. According to the system prompt, Productivity Partner GEM will be able to access and analyse the user's data from Gmail to provide productivity suggestions.
-
Superintelligence: The Ultimate Geopolitical Weapon?
By
–
Superintelligence: the ultimate geopolitical weapon? https://
youtu.be/M9F99BwiQXw?si
=se4lqdMmMKmVVDyb
… via @YouTube