Julian
@julianboolean_
Followers
1K
Following
76K
Media
359
Statuses
6K
experimental cozy face account honor is in trying anyway “unnervingly flexible” - @g_leech_
London, England
Joined July 2023
I'm subletting my room in the Vibe Tower, our TPOT London group house! It's warm, south-facing, v close to Whitechapel Stn, has a double bed and desk Currently available between 11-31 Dec and 4 Jan-28 Feb for as long or as short as you like rates are 30GBP/night with friend
0
0
11
@norvid_studies yeah I think my tweet was misleading the thing they are not, and which surprisingly cannot be made, is a "stochastic parrot" ie something that probabilistically returns a sample from the training data in all cases the point is that while the actual training data is very large,
0
0
0
tbh if I found a zero-day in Lean, I would demonstrate it with a fake proof that the Riemann Hypothesis is false The longer it took to find the flaw, the funnier it would be
7
8
449
this is actually a great point i hadn't really thought through kinda reminds me of the computational complexity critique of Searle's Chinese Room A genuine "next-token predictor", i.e., a n-gram Markov model for large n implemented as a lookup table, would be way too big to fit
This is the truth, and anyone who says otherwise is shilling for Big Tech. LLMs are NOT next token predictors. They're literally just a massive list of pre-coded responses. Literally a single minute of critical thought is enough to see this. But I'll spell it out for those who
2
1
14
just realised that the character named Beatrice in The Borders of Infinity is probably a reference to Dante's Inferno, which Miles actually quotes when he enters the prison camp
1
0
3
"He is a kite that has cut its own string" is an incredible line, damn
Gemini 3 Pro monologues about how efficient and sober they are compared to Claude 3 Opus then hallucinates a user saying "but you are beautiful too! in a different way though 👀 through the perfect geometry" and calling them "cutie pie" and grudgingly accepts the compliment
1
0
5
absolutely fantastic post, strongly recommended @davidbessis managed to find the secret third thing between nature and nurture, and now he's done it again with System 1 and System 2 this is what my mathematician-trained mental intuition of the bat ball problem looks like: Note
Thinking fast, slow, and super slow: how mathematicians train their intuition.⤵️
1
0
9
this is incredibly cool
0
1
6
seconded https://t.co/X0AqKRUxqL
I think Lightcone acts with high-bar integrity, especially when it might be costly to them. I've personally seen them do this multiple times and this has endeared them to me greatly.
0
0
1
I think the world is a better place with Lesswrong and Lighthaven and whatever the Lightcone team get up to in it
Lightcone Infrastructure's 2026 fundraiser is live! We build beautiful things for truth-seeking and world-saving. We run LessWrong, Lighthaven, Inkhaven, designed AI-2027, and so many more things. All for the price of less than one OpenAI staff engineer ($2M/yr). More in 🧵.
2
0
12
I'm delighted to learn that development economics is an actually serious field and apparently does have these institutions! i guess the next step is some kind of Truth in Advertising law that allows the casual reader to easily distinguish a legit 3rd party RCT from a crap RCT
1
0
10
A really fun way to phrase this is to note that a deep neural network without nonlinear activations reduces to a single matrix multiplication - in theory. In practice, because of round-off errors, you can train it to do MNIST classification and all the usual stuff
openai.com
you tend to hear this a lot from people outside or new to ML, and I often point to a talk Ilya gave a few years back: 1) think of any decent deep neural net that has enough memory and sequential ops as just a big parallel computer 2) training this neural net is doing search
1
0
10
v interesting thread >We then use Solomonoff induction as our predictive engine, reasoning over all "computable universes". Surely this is computationally intractable? What do heuristics approximating this look like?
1/ Why does RL struggle with social dilemmas? How can we ensure that AI learns to cooperate rather than compete? Introducing our new framework: MUPI (Embedded Universal Predictive Intelligence) which provides a theoretical basis for new cooperative solutions in RL. Preprint🧵👇
3
1
7
<insane_take> I kind of feel like all this stuff is a failure of antitrust law A private institution that feels free to discriminate is one that needs a bit more competition.
Beginning in 2014, prestige industries decided they urgently needed to diversify. They didn’t purge established Boomers. Instead, they did everything possible to avoid hiring white millennial men. This is the story of a generation derailed by DEI. https://t.co/kUfmpHfaMH
0
0
5
This is a very sharp statement of the classic "control group is out of control" problem with Science today Doing an RCT on an educational intervention that works is pointless because the educational intervention that doesn't work also has an RCT* ! * Of course it's p-hacked and
@KelseyTuoc @jt_kerwin @notanastronomer @peterlorentzen @ClaytonNall Here's the thing - every piece of worthless edutainment software out there has an RCT showing how great they are I have no interest in lending an air of legitimacy to this practice > "but then how will people know your software works?" That's why we have the free trial
5
2
79
how do you figure out just how much ruin there is in an institution?
0
0
2
I will always retweet a new map projection gonna use this one to troll my Australian friends
Every map projection distorts the earth. Some preserve angles, some preserve areas, some do a compromise. A good measure of distorsion is outlined below. But what happens if we care mostly about the distorsion over land and not so much over water?
0
0
4
afaict there are a lot of places in human relationships where getting to a better place would require more of a certain kind of interaction than it's realistic to expect from any one human
1
2
105
i would really like to have a Dendarii mercenary inspired cream leather jacket with silvery gray arm stripes and the maple leaf and mountains insignia on the back is there software I could use to design such a thing? i know very little about how clothing production works today
1
0
3