Apply for the Affine Superintelligence Alignment Seminar #shorts

Key Takeaways

Sooner or later, and I don't know, it's looking like maybe sooner, there will be AI systems that are more capable than all of humanity. It would be nice if before that happened, we developed a principled theoretical understanding of how intelligence and agency work that would allow us to be confident that those superhuman AI systems would be aligned with our interests and wouldn't end up just killing everyone. You would think that a lot of people would be working on that, but it's just an extremely technically difficult problem. Most researchers tend to decide that it's too hard and instead choose to work on easier problems with better empirical feedback loops. Or they decide there's not enough time and the best thing to do is push for an international treaty to prevent anyone building super intelligent AI until we can get some kind of handle on what the hell we're doing. And maybe we'll get lucky and AI will solve the hard problems of super intelligence alignment for us. But I really wouldn't bet everyone's lives on that. And either way, we still need people who understand this stuff thoroughly so we know what questions to ask the research AIS and how to understand the answers. If you want to help, the Aphine seminar is trying to get more people up to speed on the problem. It's running from the 28th of April to the 28th of May in our retreat center in the Czech countryside. Applications close very soon, so apply even sooner. Here's the address.