Sabitlenmiş Tweet
2025 year in review is up 🪩 cog.spin.systems/2025-in-review
On dewarping, schema inference, and embeddings
English
Louis Maddox
43.4K posts
Louis Maddox
@permutans
Combinatorially curious. Foreword Deployed Ongineer @Havelock_AI
uopuo⅂ Katılım Temmuz 2015
3.9K Takip Edilen1.9K Takipçiler
age: file encryption (Go) github.com/FiloSottile/age
rage: Rust implementation of age github.com/str4d/rage
English
Oh neat, Cloudflare R2 has SSE-C developers.cloudflare.com/r2/examples/ss…
English
Louis Maddox retweetledi
"One is about risk taking, the other is about risk mitigation" - a lot of my job as a ml eng was to reduce risk and produce impact on a frequent repeatable cadence - which is very different from open ended research
JFPuget 🇫🇷🇺🇦🇨🇦🇬🇱@JFPuget
@imrahulmaddy A researcher who codes is not an engineer still. Goals aren't the same. One is to explore, the other one is to achieve. One is about risk taking, the other is about risk mitigation. I could go on. Ways to measure their performance are not the same at all.
English
Louis Maddox retweetledi
1/7
recently i started exploring mechinterp, and i started with audio diffusion.
i spent some time cracking open stable audio 3 to see if mood has a zip code in its residual stream.
it does! it's layer 11 and you can dial it. (audio results below) 🧵
English
Louis Maddox retweetledi
@Miles_Brundage You're absolutely right — you're not going crazy
English
It is also hard to talk about it without sounding crazy
0.005 Seconds (3/694)@seconds_0
its hard to conceive of how good the models are going to be summer of 2027
English
Louis Maddox retweetledi
“AI, what color do I get from mixing black and white?”
Why do people turn to AI for simple tasks that they could easily do themselves?
In our new preprint (also to appear at CogSci 2026!), we investigate the mechanisms and dangers of people over-using AI on easy tasks.
English
Louis Maddox retweetledi
Very cool to see these conversations happening! This is what openness enables. The "tool that allows you to trace those n-grams directly to their source," is infinigram, AKA OlmoTrace from @allen_ai, created by @liujc1998. x.com/alexolegimas/s…
Alex Imas@alexolegimas
This from @TuhinChakr is brilliant. That prize winning story from Granta? Turns out it's just a bunch of random whole phrases taken directly from existing text on the internet. Tool allows you to trace those n-grams directly to their source, which is mostly random fanfiction. tuhinchakrabarty.substack.com/p/ai-slop-gran…
English
David Lyon on Donald Mackay & Claude Shannon at the Macy conferences on cybernetics (en.wikipedia.org/wiki/Macy_conf…)
English
> the monocellular quality typical of machine-generated prose
The PrettyMuchCompleterian@TLiterarian
Will Self, brilliant as ever. Ugh I hate that he articulated so much that I, many of us, was thinking. @AlannaSchubach we should talk about this!
English
FYI people, whilst this picture is pretty to look at, it is *not* the construction GPT obtained (which has far too many points to draw), this is just an example of a number field lattice that breaks for specific n.
Alvaro Lozano-Robledo@mathandcobb
Following up on the suggestion from Will Sawin, here is an illustration of the new configurations that disprove Erdos' unit distance conjecture (made with the help of ChatGPT 5.5 Thinking).
English
Louis Maddox retweetledi
Vib-Ribbon was a 1999 PS1 miracle. The game loaded entirely into the console's RAM, letting you swap the game disc for any music CD you owned.
The game analyzed your tracks to procedurally generate levels in real time.
English
Louis Maddox retweetledi
Absurd that this is well-received. Conservative, regressive nostalgia-mongering with ladels of emotional manipulation to boot. One must insist that novel forms of creativity will persevere or else nothing is being offered (Deleuze on Nietzsche: we must throw forth javelins).
G Jones@gjonesbass
every time i see a new "create a song from a single prompt" kind of product being rolled out i think about these comments from @David_Rudnick
English
Louis Maddox retweetledi
Fresh on arXiv! 😁 Our new paper reformulates tokenisation as a linear program (LP), which we solve to get SOTA tokenisers! As a bonus, this LP allows us to know how close to optimal any tokeniser is! Check it out! 👇
Jan Tempus@Jan55028368
In our new paper, we reinterpret tokenisation as a problem in high-dimensional geometry (100M dims to be precise!), which we can solve efficiently to get a globally near-optimal tokeniser! Our method consistently improves language models over BPE. See 🧵for details.
English
English
TeamPCP just did an interview where they were asked what defenders should do to stop supply chain attacks.
Their advice: pin versions to a specific hash, use least-privilege tokens, restrict IDE extensions. And then, verbatim: "The company Socket will detect the malware before the package even reaches your machine."
So... thanks, I think?
We're not putting this on the testimonials page.
But at the same time, if you're not yet using @SocketSecurity to protect your supply chain, what are you waiting for?
English
>secretary says I have a meeting with the boss
>ask if it's him or his doppelgänger
>she doesn't understand
>pull out an illustrated diagram showing the difference
>she laughs 'its a good meeting sir'
>go to the meeting
>It's his doppelgänger
English