This was the best AI Safety event I’ve ever been to. Unlike every other event like I’ve heard of, thanks to the organizers we actually spent a lot more time on trying to understand Illegible Problems in alignment, in ways I think was actually useful and something I could see being useful for actually trying to understand the problem.
I don’t think this would have happened if the organizing team didnt have people who actually understood the problem deeply (limited by how much understanding of the problem is available right now, of course).
In particular I was really really really impressed that halfway through the event there was a mandatory event where the organizers told everyone to STOP, that there was far, far, far too much focus on producing legible output over actual alignment work. I was not expecting this and I’ve not heard of any other organizers doing this for an event, ever.
Thank you so, so much to the wonderful team who made this happen.
This was the best AI Safety event I’ve ever been to. Unlike every other event like I’ve heard of, thanks to the organizers we actually spent a lot more time on trying to understand Illegible Problems in alignment, in ways I think was actually useful and something I could see being useful for actually trying to understand the problem.
I don’t think this would have happened if the organizing team didnt have people who actually understood the problem deeply (limited by how much understanding of the problem is available right now, of course).
In particular I was really really really impressed that halfway through the event there was a mandatory event where the organizers told everyone to STOP, that there was far, far, far too much focus on producing legible output over actual alignment work. I was not expecting this and I’ve not heard of any other organizers doing this for an event, ever.
Thank you so, so much to the wonderful team who made this happen.