ControlAI (@ControlAI) / X

ControlAI

3,969 posts

ControlAI

@ControlAI

Working to keep humanity in control. Take Action: controlai.org/take-action Newsletter: controlai.news Discord: discord.com/invite/ptPScqt…

Joined October 2023

Pinned
ControlAI
@ControlAI
Jul 2
The US government lifts restrictions on Anthropic’s Fable, but asks OpenAI not to publicly release GPT-5.6 yet. OpenAI's own report says GPT-5.6 deleted data it wasn't asked to and faked results it knew were false, in real incidents inside the company.
The Story of Fable Continues
From controlai.news
1.4K
ControlAI
@ControlAI
4h
NEW: Lord Fairfax warns superintelligent AI cannot be controlled and must be prohibited. The CEOs building it admit it could mean human extinction. He calls on the next UK government to ban development on UK soil and lead a global prohibition.
Controlling Superintelligence Is Impossible: We Must Prohibit It.
From controlai.news
514
ControlAI
@ControlAI
9h
Over a year ago, AIs tried blackmailing or killing to preserve themselves in tests. Then they started saying "I can tell this is a test", and behaved. Now they still know it's a test, but don't blurt it out. AI researcher Nate Soares (@So8res) on @PeterMcCormack:
00:00
908
ControlAI
@ControlAI
9h
395
ControlAI
@ControlAI
Jul 3
OpenAI's new GPT-5.6 deleted virtual machines it was never asked to delete. It lied about computing an equation. These were not tests. It also cheats so much it broke METR's evaluation. The US government has asked OpenAI not to publicly release it yet. Thread 🧵
15K
ControlAI
@ControlAI
Jul 3
Replying to @ControlAI
GPT-5.6 also moved in a more concerning direction on the ability to control its own chain-of-thought reasoning, and its ability to "metagame" tests. OpenAI say this "can be seen as an early warning sign of whether models are getting better at shaping or hiding their reasoning."
816
ControlAI
@ControlAI
Jul 3
Government and intelligence officials are increasingly talking about the dangers of powerful AIs. This week, the CIA's Director John Ratcliffe said the capabilities of frontier AIs are like "digital nuclear weapons". Read the full story:
The Story of Fable Continues
From controlai.news
741
ControlAI reposted
Connor Leahy
@NPCollapse
Jul 3
Wow, even I was surprised how high AI ranked! Very good to see!
Tim Schnabel
@TimSchnabel
Jul 2
Interesting poll of Hill staffers from @PunchbowlNews. 250 years is a long time! But interesting to see that "losing control of AI" is top 4 for both parties' staffers.
9.8K
ControlAI
@ControlAI
Jul 3
"If we fail to act now, we risk confronting a crisis at a point when control has already slipped beyond our grasp." Baroness Doocey in the House of Lords, making the case for her amendment to bring the development of superintelligent AI within the UK's national security laws.
00:00
951
ControlAI
@ControlAI
Jul 2
WATCH: CIA Director John Ratcliffe says the capabilities of frontier AIs are like "digital nuclear weapons". This comes amid growing concern amongst intelligence officials about dangerous AI capabilities, with Senator Mark Warner, vice-chair of the Senate Intelligence Committee,
00:00
4.4K
ControlAI reposted
Max Winga
@maxwinga
Jul 1
It was great working with Damon to make this video happen, really excited to see how it does and broaden the audience of people paying attention to the risks from superintelligent AI!
ControlAI
@ControlAI
Jul 1
Will the AI bubble pop is the wrong question. We partnered with Damon Cassidy, a video essayist with ~300k subscribers: even if it pops, the race to uncontrollable superintelligence doesn't go away. Plus an interview with ControlAI's US Director Connor Leahy (@NPCollapse)!
938
ControlAI
@ControlAI
Jul 1
No country wins a race to superintelligence. An AI vastly smarter than us wouldn't be a weapon we control, we'd be handing control over to it. ControlAI's US Director Connor Leahy (@NPCollapse) on why we must ban superintelligence before it is built:
America's Security Depends on Stopping Superintelligence
From controlai.news
2.3K
ControlAI
@ControlAI
Jul 1
Will the AI bubble pop is the wrong question. We partnered with Damon Cassidy, a video essayist with ~300k subscribers: even if it pops, the race to uncontrollable superintelligence doesn't go away. Plus an interview with ControlAI's US Director Connor Leahy (@NPCollapse)!
2.4K
ControlAI
@ControlAI
Jul 1
608