


Super Intelligence Policy @joinFAI


I wrote about one of the forms of surveillance that concerns me most as SI diffuses into the world: futurecraft.thefai.org/essay…
Enjoyers of gradual policy change beware, this essay compendium is about the ways we'll need to adapt to govern our country after SI starts changing everything

Enjoyers of gradual policy change beware, this essay compendium is about the ways we'll need to adapt to govern our country after SI starts changing everything
The launch party for our new compendium of post-AGI policy essays, FUTURECRAFT, is this Tuesday! Apply to attend here: futurecraft.thefai.org/


I had the chance to discuss the benefits of AI and SI with Americans for Prosperity’s Rebecca Provost on their podcast The Natural State of Things. It’s always frustrating how the benefits of technology are diffuse and the harms are concentrated, so it was great to have a chance to discuss the positives. “AI accelerating medicine” doesn’t mean an instant cure for cancer. Sometimes it means much more boring things, very day-to-day quality of life improvements like being able to ask a chatbot if some rash warrants a trip to the doctor. That’s hard to make an ad about, but multiplied across all Americans, it’s a real benefit. These are the kinds of small, diffuse gains that central planners are poorly suited to account for, and it’s why we should resist regulations that would prevent AI and SI systems from providing medical advice. open.spotify.com/show/0rHuWN…

Here's the interactive tool for visualizing the importance of national environment and candidate quality, you can see how various senate elections (both historical and future) would go if there were different national environments claude.ai/artifact/UjJ6EmBTW…

Then in 2030 Repubs can take the Senate back-- assuming Dems flipped no more than one seat in 2028, they end up at 57. If a Dem is president and the 2030 midterm is R+8.6 (the mirror image of 2018), Republicans flip 9 seats and get back to a 52-48 majority.

Look at 2028. Apply a uniform swing to 2024 results: at a tied national House vote, Republicans net 4 seats (they flip AZ, GA, NV, OH and PA; Democrats flip WI). The only way Dems flip more than one seat is if they win the national vote by 10+.

I'm hearing some folks say Republicans are locked out of the Senate for a decade. Crazy take. I can't see how the takeaway from this election is anything other than: with a blue or red enough national environment, tons of states are in play.

Actually we do in fact work on Super Intelligence policy and not cat/dog image classifiers. Did you read Sam’s announcement article? thefai.org/posts/announcing-… “AI” encompasses a lot of things that have been around since the 1950s and no I am not employed by FAI to write policies concerning spam filters
This is what it looks like when you have to embarrass yourself to align with Trump's whims.

I'm proud to work for the only DC think tank with a Super Intelligence policy team. Curious to see how long it takes for other institutions to follow suit
good morning @JoinFAI's AI Policy team is now the Super Intelligence Policy team. please don't deadname us. we have a swear jar thefai.org/posts/announcing-…

good morning @JoinFAI's AI Policy team is now the Super Intelligence Policy team. please don't deadname us. we have a swear jar thefai.org/posts/announcing-…

To make the transition from saying AI to SI as easy as possible, we are releasing an “AI → SI” chrome extension that automatically swaps "AI" for "SI" and "artificial intelligence" for "super intelligence" wherever you go on the internet! Get it here: chromewebstore.google.com/de…

Strong +1 to this
People should be careful not to have the wrong takeaways from this. Of course there are possible tail risks that would be uninsurable. This is true in many industries where we mandate insurance (nuclear, satellite, aviation, oil spills, cyber). And it doesn’t mean we shouldn’t require insurance for large but still insurable harms. The technical measures that reduce insurable risks will also reduce the odds of uninsurable harms.

I think co-insurance (for example, having them pay 5% of any catastrophic harm) might be underwritable, and it might have most of the qualities we want! (such as having them pay premiums in proportion to the risk they pose to the public).

People should be careful not to have the wrong takeaways from this. Of course there are possible tail risks that would be uninsurable. This is true in many industries where we mandate insurance (nuclear, satellite, aviation, oil spills, cyber). And it doesn’t mean we shouldn’t require insurance for large but still insurable harms. The technical measures that reduce insurable risks will also reduce the odds of uninsurable harms.
Anthropic says that even the entire insurance industry in the aggregate will not have the capacity to insure the catastrophic risk that ANTHROPIC ITSELF poses to the world lol