A green tiki mug with mint and a metal straw coming out of the top and a Bauhaus poster in the blurred background.
Posts by Aaron Steven White
Relatedly, companies aren’t just their products; they’re also pre-built networks. Why do people pay to join a frat or go to business school?
I'm reading Anthropic's recent antics as them knowing that the individual coding agent plan is dead as most of us will be using local models for day-to-day engineering in 12-24 months, and shifting their revenue towards big enterprisey Glasswing-style shenanigans.
Fleshed this idea out a bit.
Love when you bounce a track and all the inserts are just gone.
Claude Code has certainly made me write code faster. But it turns out, the bottleneck is still experiment run time and experiment analysis
tfw @aaronstevenwhite.io brings an analysis as sharp as a knife to your half-baked Saturday-morning thoughts: aaronstevenwhite.leaflet.pub/3miwsz2hdv22i 🤯
If we're going to own our data, let's actually own our data.
Which is to say: No, really, y'all, we're doing this. 💖🧠
Fleshed this idea out a bit.
I am not one of those who believes that history repeats, but consider the following from Herodotus: the Oracle at Delphi told the fabulously wealthy King Croesus that if he invaded Persia he would "destroy a great empire." Croesus invaded and lost--the empire he destroyed was his own.
Realizing that some of my inexplicable sadness about the Artemis II stuff is that it feels like a trans-dimensional communication from an America that took a different path and chose joy instead of fear and hate.
An interface for designing dependent optics in panproto.
lol.
You know what would be so sick: a Max-like interface for defining (meta)schema/data transformations.
I mean. This isn’t really just a challenge in the context of propositional attitudes. It’s relevant to any structured representation of actions, events, states and their participants.
And if you do allow this kind of dependent content type, which I think you have to, you end up needing a generalization of lenses to dependent optics, so you can scope your content translation correctly.
The challenge there is that the structure of an attitude content is itself dependent on the attitude type: intents, preferences, beliefs, and knowledge plausibly take distinct content types (at least if you take the way we linguistically realize these contents to be reflective of their structure).
Seems crucial that any such generalization take the same open domain approach to typing attitudes themselves (not least because attitude ascriptions can be the contents of other attitudes).
One kind of interesting consequence of generalizing this idea to arbitrary attitudes is that you end up with an explicit representation of where the contents of two attitudes are incommensurate and how big the difference is.
Relatedly.
Specifically about intents, but what should machine-readable propositional attitude ascriptions look like more generally? (This shows up also in the context of, say, linking claims to evidence, but I suspect there's many more kinds of attitude-{attitude, world} relations one might want to capture.)
Also the book that got me making oleo saccharum on the regular.
The book that finally got me acid adjusting religiously.
There’s been a glitch in the simulation. You have been sent back to April 4, 2026
Will write a thread or two this week on the basic idea if I have some time.
Had Jane Chandlee here yesterday and it was awesome.
chandlee.sites.haverford.edu
Really inspiring me to get back to my long-running side project looking at the subregular hierarchy through the lens of spectral learning.
I don’t know, man.
Welp.
My talk from Atmosphere wasn't available, so I used the Power of the Atmosphere to build a place you can see it: ionosphere.tv/talks/rj8Xv62
You can also leave comments/reactions if you like, and the rest of the talks are (mostly) also there. Explore!
Piracy for me but not for thee.
hyperdeclarative.