Joined June 2013
58 Photos and videos
Justis Mills retweeted
man: i wish to publish reviewer 2: your paper is no good man: i'll do anything to improve reviewer 2: it's simple. you must read the work of the great scientist Pagliarini man: *bursts into tears* but i am Pagliarini
a first: in rejecting an article I submitted to a journal, reviewer 2 noted I failed to engage the work of one Andre Pagliarini
88
5,369
73,010
1,932,613
Justis Mills retweeted
wu wei....otherwise known as lao tzu faire
7
21
139
5,126
Justis Mills retweeted
I told him he can't eat the street lamp and now he has been sulking for 20 minutes
107
6,529
77,220
924,744
Justis Mills retweeted
Scott Aaronson has, for the first time, put out a paper in which a key technical step in the proof of the main result came from AI. He describes his process using GPT5-Thinking. 'There's not the slightest doubt that, if a student had given it to me, I would've called it clever'
Yet more evidence that a pretty major shift is happening, this time by Scott Aaronson scottaaronson.blog/?p=9183&f…
38
115
1,300
384,110
Justis Mills retweeted
According to the IMF, world GDP per capita (PPP) is now $25,591. That's pretty incredible. That's about where the United States was in 1972. What a victory for humanity.
79
243
2,978
246,077
more like for want of fault tolerance the kingdom was lost, smh
31
People in the EA social sphere 5 years ago: "what? this person doesn't believe in the importance of the long term future?" People in the EA social sphere now: "what? this person believes there will be a long term future?"
47
Justis Mills retweeted
Replying to @KelseyTuoc
"i'm gonna MAKE these objects permanent!"
1
23
271
It's really beautiful how almost all grandparents occasionally call their grandchildren by their parents' names. The memory is so deep - you really are always their baby!
1
38
When I think of being immortal, like in an idle fantasizing way, math and games suddenly seem way more appealing. It would be fun to get really good at Warcraft 3 and Galois theory, if time was non-scarce. I guess because the process of getting good at those things is itself fun.
45
Something makes me sad about o3's suggestions for like, communication. Like I'll run an email by it for typos or tone mismatch or whatever and 10% of its suggestions are straightforwardly good, but a lot are like... turning the email into the Most Typical Version of itself. Sad.
1
1
45
I mean that's softmaxing baby! And also, relatedly, that's RL more generally, babeyyy. But I don't think custom instructions/temperature/base model fixes this. You can push it into another basin, but it's always totalizing. Goals ain't everything, and good ones can be nuanced.
1
37
Been trying to build some infrastructure around my behavior recently, with the dorky metaphor of it being like a mech. It does feel kind of mech-like, though! Less nimble, more powerful, easier to get where I want to go faster, easier to miss details. Requires a lot of focus.
31
Feel like we're in a weird place with terms of service/privacy policies where for a long time it was impractical to read or distill them and so necessary via triage to just click through, but now LLMs do let you get a good first approx so I feel a weak inner obligation to look.
1
36
It can't be retroactive, though! There are too many contracts. Weird how new affordances like, enable patterns of behavior that retroactively make previous patterns of behavior seem reckless or vaguely wrong. Makes me appreciate the value of amnesty, and looking ahead.
31
New novella, title Crossing the Wetlands! Link in reply. Once sentence summary: a druid helps a mysterious traveler cross the wetlands, paid in part by the promise of sex. Cozy vibes, nsfw but not (in my judgment) graphic.
1
58
justisdevanmills.com/crossin… If you read it, I'd be curious to hear your thoughts!

27
Justis Mills retweeted
10 Apr 2025
it’s an anti abundance agenda and I probably won’t see eye to eye with it. i see the magic of America as whatever feeling is encapsulated in Ferris Bueller’s day off - the almost rigorous pursuit of fun, always getting away with it. Doubling down on your monopolies
6
4
89
9,422
I wondered what would happen if I asked ChatGPT to do the same thing, but to remember that it can't experience nociception during inference. Which I think is true, basically! There's no negative reinforcement anymore at inference time. Leading the witness? Absolutely. Anyway:
28 Mar 2025
I asked chatgpt's new image model to script and generate a series of comics starring itself as the main character. The results genuinely gave me chills. I'll post them all in a thread below.
1
55
And also:
1
16
I don't really have any commentary to offer. Left an an exercise for the reader.
17