A common mistake I see people make is that they assume AI risk discourse is like the left image, when it's actually like the right image. I think part of the confusion comes from the fact that the upper right quadrant is ~empty. People really want some group to be upper-right.

Works pretty good for a “partially assembled” engine :)

It's fascinating that American higher education has decided on autobiographical fiction as a means to apportion one of the world's most valuable resources.

Matches my experience. I would not have said this two years ago but now I think it is not just true but important.

AIs making images of guns is not on my most remote threat list. It is very hard to hurt someone using an image of a gun. Corporations that prevent this do so to protect their brand safety - which is usually what "safety" _means_ to a corporation.

So freaking proud of the AGI safety&alignment team -- read here a retrospective of the work over the past 1.5 years across frontier safety, oversight, interpretability, and more. Onwards! alignmentforum.org/posts/79BPxvSs…

We’re open sourcing Vivaria, our platform for running LLM agent evaluations and conducting research in eliciting AI capabilities. 🪴

Want to know what SB 1047 actually says*, but don't have time for a full read? Zvi Mowshowitz delivers us this summary (linked in comments) *without all the biased exaggeration from a16z!

AI corporations complained, got most of what they wanted, but they’re still shrieking about bill SB 1047 just as loudly as before. 1) "Off switch requirement kills open source!" ➡️ has an open source exemption. 2) "You can't guarantee no critical harms!" ➡️ it's now reasonable

Was recently debating some friends about how “purist” AGI safety proponents should be. One key reason I want high standards: as the number of people justifying their proposals by appealing to AGI safety grows massively, the big bottleneck will be figuring out who’s right.

productivity growth is 3.4% "we had averaged just 1.1 percent in the decade prior to the pandemic"

New post about safety cases at AISI! To complement to our empirical evaluations of frontier AI models, AISI is planning collaborations and research projects sketching safety cases for more advanced models than exist today, focusing on risks from loss of control and autonomy.

Boeing was paid billions more than SpaceX to be seven years late on a mission it could not complete . . . And will now be rescued from total disaster by SpaceX Never forget, many bureaucrats wanted to sole source all of this to Boeing.

Everyone's talking about Sakana's AI scientist. But no-one's answering the big question: is its output good? I spent hours reading its generated papers and research logs. Read on to find out x.com/SakanaAILabs/s…

I'm into feminist transhumanism. Nature is sexist and technology can balance the scales. All hail the pill, abortion, C-sections, tampons, baby formula, antibiotics, egg freezing + IVF + surrogacy, antidepressants, prenatal screening, epidurals, and one day artificial wombs.

Just going to say it again, AB 3211 is a considerably worse bill than SB 1047, and has not been red-teamed in public nearly as much as 1047. Whatever you think about 1047 (I’m still not a fan), go read AB 3211 or Zvi Mowshowitz’s analysis, Brian Chau (SF 08/18-09/03)’s, or my own (tho mine’s dated).

David Dayen This is how the CalChamber described SB 1047! They managed to get Politico to cover it in their CA newsletter ahead of the final votes, without releasing the full results, which found their way to me...

I resonate with this principle from Dean Ball's post last week and AI Snake Oil's post last month. Agree freedom+tech are both usually great. I'd be pissed about limits on tech *I* like based on speculative risks *I* don't buy. But we're in a tough epistemic bind 🧵

I have massive respect for William Saunders, friend and former OpenAI colleague. Here's a long-ass blog post by him with his reflections on the AGI race: williamrsaunders.substack.com/p/principles-f…

We thought it would be helpful to have all of the similar themes/components from each of Deepmind's Frontier Safety Framework, OpenAI's Preparedness Framework, and Anthropic's Responsible Scaling Policy, in one place.