is/ought retweeted
If you have an UTM, it can simulate any other UTM. Make nodes UTMs, make edges x->y with weight w(x)/2^2x(y), where x(y) is the description length of y in x, and w(x) is one over the sum of this over all nodes. Now you have a markov chain.
This object is in some sense canonical. But it doesn't have a canonical index. But there is a canonical index of every UTM from every other UTM. So there is a meaningful notion of relative location.
Anyways. I have Opus try to make a visual novel out of this idea, partly based on Eliezer's "the finale of the ultimate meta mega crossover". It works on this for like 5 hours.
But then it starts play testing it (it made it some kind of game rather than just a comic / visual novel). And now its just been playing the game for like 2 hours
the only good outcome here—benevolent ASI dictator—is made less likely, not more likely, by pauses and gov't/normie attention being paid to ASI.
unfortunately, i think the outcomes of this week—the massively increased attention being paid to ASI by gov't/normies—is going to be net-bad. I think it's possible it will slightly reduce x-risk, but i think it has significantly increased the risk of hyper-centralizing ASI.
the best outcomes all involve human disempowerment. the worst outcomes all involve human disempowerment.
as a pro-ASI person, i feel bad for the doomers. they say: "ASI is dangerous" and the people debating them essentially respond "gpt-5 is perfectly safe".
the doomers should specify (and repeat), in every convo: "here's what ASI is. we're only debating ASI. focus on ASI!".
There's this weird gambit many in tech play where they frame themselves as excited about ASI, when, actually, they just don't believe ASI is going to happen. What they're excited about is instead merely a useful, but not at all superhuman, AI.
people get Jevon's Paradox with regard to ASI wrong. Jevon's Paradox/basic logic means demand for compute will be infinite once ASI is achieved. demand for humans will go to zero, however.
