I feel like I'm the only one that thinks this Dyson sphere method is a little dubious. What system is going to be used to collect energy using the captors and send it to Mercury? How will it be received on Mercury? The total power collected toward the end is more than ${10}^{24}$ W. If whatever process is used to disassemble the planet is 90% efficient, the temperature required to radiate the waste heat over Mercury's surface area is about 7000K. This is hotter than the surface of the sun, and more than twice the boiling point of both iron and silica. In order to keep this temperature below the boiling point of silica, we would either need the process to be better than 99.98% efficient, to attach Mercury to a heat sink may times the size of Jupiter, or to limit power to about ${10}^{20}$ W. If melting the planet isn't our style, we need to limit power to about ${10}^{19}$ W.

Doesn't it mean that the last stage of the process takes a few orders of magnitude longer, not the whole process? The process consists of a series of N doublings. For all but the last... 7 or so doublings, the waste heat can be dumped into Mercury itself. Only for those last doublings does waste heat start to overheat Mercury. Right? So it's just that those last 7 doublings or so need to slow down (or rather, stop growing exponentially)

More interesting proposal: Maybe actually a better strategy would be to just deliberately overheat Mercury at that point, turn it into an expanding cloud of superhot gaseous material, and then scoop up said material as it cools down? Not sure if that's possible, maybe the cloud wouldn't expand enough.

Another interesting proposal: Smash Mercury to bits by colliding other objects like Ceres into it. 

Another, more practical proposal: Once Mercury is saturated (can't double any more without overheating it) switch to mining those cold asteroids out in the asteroid belt, with your giant fleet of spaceships you've built with the infrastructure that blankets Mercury and orbits the Sun. The asteroids are all spread out, so it's easier for them to radiate heat... right?

Just spitballing here.

The author is "Ellis Elms." Hmm.

I made this post because I wanted to link to the image on substack and couldn't put the image in there directly.

But yeah, the takeaway I got from this toy model is that the time horizon will be exponential until the last couple of doublings before it goes infinite. (Well, that's what happens if half the progress is coming from imrpoved intercept and half from improved slope. The more it comes from improved slope, the more straightforwardly superexponential it is. In the real world I don't think we have much of a clue yet where the progress is coming from.)

If you feel like I am falling short of your ideals, I'd be interested to hear it. I'm trying to live up to them basically. it's not my top priority, i'm not trying hard, but it matters to me.

Related previous work: https://www.lesswrong.com/posts/QEBFZtP64DdhjE3Sz/self-awareness-taxonomy-and-eval-suite-proposal

Summary of previous work below:

Huh, I think continual learning would be a pretty big deal w.r.t. AI danger. Can you say more about why it wouldn't?  Seems like it would dramatically increase horizon lengths for example.

Back when many expected takeoff in 2027 or so, it was pretty reasonable to assume that the probability of a conflict entirely unrelated to AI was low.^[3] But the forecasters behind AI 2027 now expect takeoff in the 2030s.

I'm saddened that this is the takeaway from our new model! It seems misleading. Here is a graph of my+Eli's timelines over time:

https://x.com/DKokotajlo/status/1992316620254155028