I've been thinking more about the lines of code metric recently. I initially shared Nate's intuition that your post seems to overindex on lines of code (quoted below, from your ftn 1), as I recall telling you at the time.
Directionally, I disagree with the size of the update you've made here based on a single metric - noting the size of the update you've made comes through more in our brief conversations than in this post. I have a general take of 'in the early days of RSI, no single metric is going to tell us what's going on fully' position, and I feel li
I did a lot of BP debate in college and its main effect was taking up time I would have otherwise spent doing much more valuable EA and AI safety work. I don’t think it helped or hurt my epistemics.
(This story was written in collaboration with Claude. It's not intended to be realistic, but to spark interesting ideas.)
I enjoyed the story! How did you use Claude to help write this? I’m surprised that it was capable enough to meaningfully assist; maybe I should try using it for fiction writing uplift.
A possibility that occurs to me: If early automated researchers refuse to work on capabilities, then an irresponsible AI developer could use low stakes control to prevent them from sandbagging on capabilities work. Maybe the use of low stakes control to circumvent this refusal is a significant knock against further research developing low stakes control methods.
(As it stands, I don't take this point as a substantial update against low stakes control research because it doesn't seem likely enough that early automated researchers will in fact refuse to... (read more)
I've been thinking more about the lines of code metric recently. I initially shared Nate's intuition that your post seems to overindex on lines of code (quoted below, from your ftn 1), as I recall telling you at the time.
... (read more)