This is a special post for quick takes by Simon Lermen. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
I recently joined Inkhaven for the month of November. Inkhaven is a program run by Lighthaven in Berkeley where people for one month are supposed to write one blog post every single day of the month. The idea was inspired by Scott Alexander—that if you blog every single day and are consistent at it, you're going to get quite far according to him. Inkhaven takes place in Lighthaven which is hosting many efforts dedicated to AI safety.
I myself wanted to get better at communication for AI safety, and it seemed like a great opportunity. I don’t have a lot of experience blogging or communicating, I have a few blog posts mostly on technical projects I have done but tried to steer more into opinion and position posts in the future.
And I did have the feeling that I wanted to talk about issues in AI safety. Especially recently, I felt that there was a strong split in the AI safety community and I hoped that I could get my voice out there. One of the things that motivated me to take part in Inkhaven was the divided reaction to the book that Eliezer and Nate published. I had always thought that they and especially Eliezer made very straightforward points and had these very intuitive deep insights into AI safety. So I was saddened when I saw many people in the AI safety community attack the book with in my view usually misguided arguments.
For example, Will MacAskill wrote a review with a very aggressive response to Eliezer. One of the things he pointed out was about Discontinuities. He took a single sentence from the book in the fictional chapter on the takeover scenario: “A new sort of mind begins to think.” His reply sounded as if he had found out a huge error in the book. Will misrepresent their position as if it relies on fast take off, then misrepresents that the section describes fast take off and then falsely asserts that fast take off is so unlikely the book has a major flaw here. And he didn't really present any argument for why it is unlikely.
I felt a strong feeling that Eliezer and Nate had created a very clear overview of the dangers of AI. But then so many people from this community went against them. And I tried to push back on some things, but it just left me with the feeling that I wanted to get my voice out there more. If it matters, I want to be able to really speak up and make a point.
I have also been reading Scott Aaronson’s blog and his other writing for quite some time. I really enjoyed his book quite a long time ago. He and the other supporting writers are also big reasons why I wanted to be at Inkhaven. I also sometimes felt that I had a lot of ideas and I wanted the ability to write them down. For myself, writing makes me think better about topics. Furthermore, I'd be very glad if I can successfully reach some people with my writing. And if I'd be able to very quickly post responses and to quickly make write-ups of smaller research projects. My very first post at inkhaven talked about the ethical consequences of uploading humans and what it means for the world if open-source models could have subjective experience. I'm very happy for any feedback and I hope I can make a lot more approachable posts on topics like misuse or AI relationships with humans in the upcoming days.