Pseudonym
Joel Tau
The universe only needs to resemble our intuition locally, because our intuition comes from the part of the universe we are in. On different scales or in different places, there is no need for it to seem sensible.
Perhaps it is that simple.
GLM 5.3 is scary. It appears to have relatively strong cyber capabilities: not Mythos level, but it seems like it could cause significant harm.
Where is the focus on self-improvement, combating akrasia, productivity, and rational action? These things are important, and we know they’re important, but I don’t see much work on them recently. Practical posts seem to be doing this, but they don’t feel like enough. There isn’t common knowledge here about this like there is with rational belief.
Without categorization, it’s difficult to describe what has and what has not already been written about. So, let’s categorize them temporarily. Split the art into self-therapy[1], productivity (getting things done once you have motivation), and dealing with akrasia/getting motivation[2]. Productivity and dealing with akrasia appear to have been addressed significantly already, so I think therapy deserves the most focus.
I expect rationality to help with these things because good epistemics could allow us to filter through the massive memeplex of self-help. Unfortunately, it is very difficult to get good information on therapy. Rationality doesn’t solve that problem, but it seems like it should allow for extraction of at least some useful information.
Or perhaps it’s relatively simple and large swaths of systematic self improvement come from character traits, trying very hard, getting help from other people, and compound interest on continual improvement. Increased organization of this information could be quite helpful.
Either way, this has probably been addressed about thirty times, and I’m missing at least some significant self-improvement writing here.
Joel Tau’s Shortform
Rationalism is not more popular because it’s not especially memetic. This is partly because “many people are not rationality-shaped”, as Gordon said, partly because of the length of expositions, but it goes beyond that.
Additional factors:
Unlike religion, politics, ideology, etc. it usually doesn’t stir up strong emotions. I think that most people aren’t really interested in improving thinking in the way that they are interested in other things. There are rarely social reasons, monetary reasons, or emotional reasons.
People aren’t sharing rationalism much (or at least, not many people come to LW because of people sharing it). Looking at the 2023 survey results, only % of people are referred to LW by friends. The expected number of people that each person directs to LW is probably well below 1, so exponential growth doesn’t work effectively. People usually up learning about rationality through the internet, HPMOR, ACX, etc. as the survey indicates.
In this case, we’d expect to see gains because this is adjacent to tasks that are easy to verify. I expect that we’d see less gains on tasks which are a lot harder to verify (this precisely matches my experience). Even still, there’ll be gain on those from general capabilities increases. The model of “progress will happen adjacent to where there are verifiable rewards” remains.
There could be some build-up of capabilities like you explain here (where we don’t see it until it’s here), but it’s difficult to see how that would extend into areas like research taste within the current paradigm.