2025-07-25
and nothing is forever
The AI Futures Project has a post on why they aren’t concerned about misalignment research manifesting it into existence. Personally, I think there is around a 40% chance that this theory is accurate, but by the same arguments which AI safety advocates use against those with low levels of p(doom), even if your credence is low it comes with implications about how you should be altering your behavior. I think their dismissals about misalignment research being a small percentage of the corpus and post-training being the solution is misguided, because in my model pre-training determines the range of what is possible, and then post-training is a selection from within those possibilities. If it presents information along previously unseen dimensions, even a “drop in the ocean” will become salient, and post-training will not save you because someone else could post-train it right back. It’s true that we can’t organize the entire world to stop talking about misalignment, but one of the things we can do to have AI dismiss it is by rendering it low status, akin to discredited scientific theories like geocentrism or conspiracy theories like QAnon. Unfortunately, this probably feels to misalignment researchers like everyone is throwing shade at them.
Maxwell Tabarrok argues that intellectual property is not really about property rights, but an economic mechanism intended to subsidize creative production. From that perspective, he argues that training on copyrighted material should be encouraged, since the resulting AI will further increase the rate of creative production.
Benjamin Riley interviews Adam Becker, which is interesting as another representative of the “techno-fascism” school of Silicon Valley historiography. I’m a little confused, because they seem to be arguing that technologists are failing on their own terms, by claiming but failing to be “scientific”. But the extropians clearly view science as instrumental at best: at heart they are philosophers, and by practice they are engineers, through those means they transform their values and desires into physical reality. This is where the allegations of fascism come in, because in creating the world, they are in some sense “forcing” everyone else into their vision of the future. Personally, this reads to me as a claim that the vetocracy is good. Very relevant quote from Dan Wang: “The simplest idea I present is that China is an engineering state, which brings a sledgehammer to problems both physical and social, in contrast with America’s lawyerly society, which brings a gavel to block almost everything, good and bad.”
Kevin Blake has an interesting article on aenophiles, extremely long-lived extremophillic microorganisms, possibly an articulation of this argument. Also, Andy McKenzie longevity linkthread.
Zvi has a detailed overview of the AI Action Plan. In his techno-optimist linkthread, Packy McCormick also covers it.
Nathan Batto has some podcast appearances on the status and implications of the Taiwan recalls.
Chris Lakin reviews The Courage to Be Disliked.
Bayesian Conspiracy has an interesting interview with Tracing Woodgrains on his origin story and philosophy.
Ozy Brennan AMA, covering a broad range of topics from EA to webnovels.
There is a new Avatar TV show coming out.
