Waiting for AI to demonstrate dangerous advanced capabilities to take it seriously is like waiting for a dam to burst before moving to higher ground when you already know the cracks are there.

If you have money, invest it now in building strong AI governance and international coordination. Without it, your wealth might not matter much in the not-so-distant future.

One of the most frustrating things in AI governance discussions is when decision-makers fake understanding instead of asking questions. It’s not about intelligence (some are smart) but breaking that cycle of surface-level perspectives could unlock so much progress.

Thinking an ASI will be aligned by default because today’s models seem fine is like predicting a tsunami’s behavior by watching waves in a bath: the basic principles are the same, but the scale fundamentally changes the phenomenon.

Assuming we’ll control future techno because we’ve controlled past ones is flawed: it generalizes from successes while ignoring that, if control had been lost entirely, we wouldn’t be here to observe it. It’s like declaring yourself unsinkable because you’ve never hit an iceberg.

If you think "there’s no alternative but to race for building AGI as soon-as-possible because rivals will do it if we don’t," you’re probably confusing inevitability with a lack of imagination.

A few months ago, a colleague told me "AI posed no systemic risks". Now, it’s obvious to anyone who thinks for 30 seconds. Funny how people who don’t want to see a problem will make up stories and actually believe them. (See assets.publishing.service.gov.uk/media/679a0c48…)

True. 👇

When will AI systems be able to carry out long projects independently? In new research, we find a kind of “Moore’s Law for AI agents”: the length of tasks that AIs can do is doubling about every 7 months.

nice analogy from Jack Clark newsletter this week

Things are moving fast. Coordination and control guarantee should be the top 1 priority.

Ursula von der Leyen: "When the current budget was negotiated, we thought AI would only approach human reasoning around 2050. Now we expect this to happen already next year." ec.europa.eu/commission/pre…

President of Singapore speaks of AGI and the race dynamic👇🏻 istana.gov.sg/Newsroom/Speec…

AGI UNCPGA Report uncpga.world/agi-uncpga-rep…

In his first official address to cardinals, the Pope warned of the dangers of AI to “human dignity, justice and labor.” politico.eu/article/pope-l…

The US Select Committee on the CCP realize the stakes related to AGI and the lack of control over these models. youtube.com/live/GDNrUZBZD…

📟 We show OpenAI o3 can autonomously breach a simulated corporate network. Our agent broke into three connected machines, moving deeper into the network until it reached the most protected server and extracted sensitive system data.

“If you’re going to work on export controls, make sure your boss is prepared to have your back,” one staffer told me. For months, I’ve heard about widespread fear among think tank researchers who publish work against NVIDIA’s interests. Here’s what I’ve learned:🧵

We just published a new paper 👇🏻