Claude just keeps getting closer and closer to being Multivac and I keep thinking about this Asimov short story blog.ac-versailles.fr/villaroylit/...

Claude just keeps getting closer and closer to being Multivac and I keep thinking about this Asimov short story blog.ac-versailles.fr/villaroylit/...

This is incredible, amazing work

Would love this so much but he leans on it as an argument still so doubtful

Interestingly these tokens probably predict your response fairly well. Did you remember what you were going to say?

Well, it was speculated that Yudkowsky threw a basilisk at an opponent to win at least one of his AI box experiments. And immediately banning Roko has the flavour of "you figured this thing out that I figured out earlier but it was supposed to be secret!". But this is pretty weak I will admit lol

Presumably only a relatively small fraction of cars are using Google maps at any one time, particularly during rush hour because it's usually just the same route for people every day. But for self driving cars they would always be using navigation software so load balancing maybe even better

This is a great point. Another mitigating factor: if the deadheading cars are being load balanced dynamically you might just use existing roads more efficiently so overall throughout increases a lot. Google already does this with maps iirc

Yeah I think this is right. Unless people actually start sharing rides more, total cars on the road can only increase. Any amount of deadheading from personal vehicles is pure increased traffic. Doubling seems high but definitely possible

Yeah I think it was the first thing he's written that I gave up like a quarter of the way in. Feels weird to me because the Law post from right before it was really good imo

The simcluster is tightly bound along many dimensions, including many that were not directly selected for (like what posts we find interesting). But why is it so strongly 'cat' on the cat-dog axis?

Yes, but in defense of the other side, have you ever had a technical conversation with someone (in something you are expert at) and realized after the fact that you said something dumb when your mind was navigating something higher level? Because I definitely have :/

For the particular study above, primary statin cessation due to side effects was the main inclusion criterion, so it includes only these people deliberately. Their initial massive confidence intervals on their prespecified methodology I think backs up my intuition here

So if you design your trial to pull all of these people, plus other less-sensitive people who had 'more severe' (loaded term) side effects, the more sensitive ones end up with higher variance overall, masking effects that trials should otherwise be powered to find

Another difficulty in acquiring reliable data on side effects is that there are patients that experience more side effects in general, are more sensitive to stimuli of all kinds, and some that have mental factors like neuroticism or anxiety that worsen experienced side effects

Fascinating, I agree with Joe that the side effects you describe were by far the main downside when I learned about this in medical school. I think your reply here is some of the reason

It's a little uncharitable to crop out his reply, which frames it differently imo

Excellent post, thanks!

Elastic Energy Recovery in Animal Locomotion Compare how 8 species recover mechanical energy through tendon elasticity during movement. Interactive chart with biomechanics research and citations.

Chart generated via Claude, because apparently the one I remembered seeing years ago doesn't exist and was my own hallucinated memory. Caveats apply for potential for Claude's own hallucinations here though I think it sourced things really well
claude.ai/public/artif...

We have nothing on red kangaroos however, at 50-70%. Their oxygen consumption barely increases with hopping speed
spot.colorado.edu/~kram/kangar...

While running, tendons, ligaments and muscles recover energy generated through active muscle contraction. Humans do pretty well! The bulk of it is from the ankle plantarflexors through the Achilles tendon.

Tell funny jokes. This feels like it should have fallen months or even years ago but is apparently really difficult for them

Alternate: account survival time after posting "I've decided to leave this app" or equivalent

I like it a lot, oddly it feels more surreal than your 'surreal' ones

This is really cool!

The Bluesky one is too new and just has that but for example, this one from Twitter

'Magellanic circumcogitation' getting saved to the brand-new Bluesky "great norvid tweets" folder

Or is it more like we keep to the useful, applied math edges, and the great gods roam the vast uncharted-to-humans oceans of pure math

What does this look like to humans, in practice? We ask Claude code for something, and it implements a "Fermat-Shannon capacity transform with Codex 7.34 optimization", and we just accept the miraculous mathematics with no chance of understanding?

Shaking things up for 2026 🙂