Sign in

Burny

@burnytech.bsky.social
881 followers 7.9K following 152 posts

On the quest to understand the fundamental mathematics of intelligence and of the universe with curiosity. burnyverse.com Upskilling @StanfordOnline

PostsRepliesMedia
Reposted by Burny
Yuxi on the Wired @yuxi.ml · 05/06/2026
a technological phase change in chemistry before 1938, thermometer, years to determine a structure after 1950, spectroscopies (many types), hours to determine a structure
1153
Reposted by Burny
Matthew Green @matthewdgreen.bsky.social · 08/10/2026
If you haven’t asked abliterated GLM 5.3 to hack the Internet’s core routers, then you’re a happier person than I am right now.
214619
Burny @burnytech.bsky.social · 27/09/2026
Explosion of AI generated mathematics with counterexamples to conjectures and the future of AI mathematics with Artificial Grothendieck Intelligence I will look at the two biggest highlights: the Erdős Unit distance conjecture and the Jacobian conjecture + Artificial Grothendieck Intelligence
220
Reposted by Burny
Singularity's Bounty e/CC @catblanketflower.yuwakisa.com · 21/09/2026
The philosophers have ruled that empirical evidence on LLMs is disqualified Of course a philosopher would hold that view
162
Reposted by Burny
Ant Stanley @iamstan.dev · 06/09/2026
claude.md as a symlink of agents.md is the way
0221
Reposted by Burny
mr. TIM @timkellogg.me · 05/09/2026
Astra is *very* good at controlling robotics this was an emergent ability through the last several generations, not because they were explicitly trained for robotics, but because they had a lot of real-world data in pretraining openai.robocurve.org/gpt-6-astra/
openai.robocurve.org
GPT-6 Astra on robotic manipulation
OpenAI's GPT-6 Astra vs Claude Fable 5.1 controlling a pair of YAM arms under the same agent policy: 19/20 vs 8/20 on block-into-bowl in interleaved blinded pairs, 2/20 vs 2/20 on the puzzle, with 80%...
5555
Reposted by Burny
SE Gyges @segyges.bsky.social · 06/09/2026
i usually have an llm running when i am here. people are telling me the thing i am currently doing does not happen. it (additionally) is like having someone tell me that spotify doesn't work
31738
Reposted by Burny
mr. TIM @timkellogg.me · 05/09/2026
three axes of test-time compute: 1. reasoning 2. agents 3. latent space in looped transformers Astra seems to be able to seamlessly tradeoff CoT for latent space
François Chollet v @fchollet
X.com
Seems like test time scaling has gained a 3rd axis: latent space reasoning iterations in looped transformers.
François Chollet • @fchollet. 4d
Test-time scaling has two axes: running agents over longer timeframes (depth), and running a larger number of agents (breadth).
Everybody knows about the first axis, but the sec...
4274
Reposted by Burny
Thorne 🌸 @ens0.me · 05/09/2026
This is a good starter pack. You should spread it around as much as possible, and I'm not just saying that because it keeps getting me new followers. bsky.app/starter-pack...
410614
Reposted by Burny
Evan Greer @evangreer.bsky.social · 04/09/2026
Age verification makes everyone less safe. Age verification makes everyone less safe. Age verification makes everyone less safe. Age verification makes everyone less safe. Age verification makes everyone less safe. Age verification makes everyone less safe.
techdirt.com
Hackers Had A Live Feed Of Every ID This Verification Company Scanned. For Over A Year.
From the very beginning of this recent obsession with identifying everyone online (yes, they like to call it "age" verification, but it always ends up as identity verification), we've been pointing out that it was a huge privacy nightmare waiting to happen. Or maybe it wasn't waiting. Maybe it was already happening. This week a...
81148287739
Burny @burnytech.bsky.social · 04/09/2026
Claude formalized proof of Fermat's Last Theorem! Nice! Next step is formalizing full Weil conjectures proofs and Inter-universal Teichmüller theory!
120
Reposted by Burny
Jen , the MT liberal bird lady @liberal-jen.bsky.social · 03/09/2026
A murmuration of starlings above Sardinia, Italy. One of my favorite natural phenomena.
22280621666
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
These models are also deceptively elegant in important ways. Their complexity largely reflects the complexity of their environment. The fact that they can pull this off with “just a bunch of matrix multiplications” is precisely what makes them so relevant for human neuroscience.
161
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
Finally, we argue that the apparent complexity of LLMs should not discourage us from trying to understand them. They are not black boxes! They are fully transparent—and we hope the explosion of interpretability work continues to cross-pollinate with computational neuroscience.
161
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
In LLMs, syntax and semantics are processed through the same neural mechanisms and both are encoded geometrically in a neural population code. If the brain follows a similar strategy, we should not be surprised that syntax/semantics are co-localized and functionally entangled.
131
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
For example, years of neuroimaging work tried to disentangle the cortical bases of syntax and semantics based on their separability according to linguistic theory—but more recent work indicates that syntax and semantics are highly coextensive throughout the language network. Why?
141
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
Later in the paper, we walk through some examples of empirical work in speech and language neuroscience where LLMs might provide some insights 🧠
131
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
To the LLM, these are all just patterns in context—it will learn them insofar as they are *useful* for producing natural language. This context-first approach to language renders different types of linguistic structure learnable by a simple, general-purpose learning algorithm.
121
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
Certain patterns are highly regular; these are the kinds of patterns that are more easily detected and have been richly described by linguists. Other patterns are more abstract, graded, or multidimensional, and therefore harder to describe.
121
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
This dense, geometric representational format also yields a surprisingly powerful form of generalization: LLMs can interpret (and produce) novel linguistic contexts by interpolating the meaning of new locations in this geometric landscape (within the bounds of their prior experience).
131
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
Principle 1 has important implications: for example, it helps us understand how many seemingly different structures of language (e.g., syntax trees, semantic categories) can be unified in a geometric lingua franca that supports neural computation.
132
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
Principle 2: They replace rule-based learning with a generic, self-supervised, context-driven learning algorithm for reproducing the statistical structure of real-world language.
131
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
Principle 1: They encode all of the structures of natural language (syntax, semantics, pragmatics, etc.) in the geometry of a continuous, high-dimensional embedding space—i.e., a neural population code.
121
Reposted by Burny
Sam Nastase @samnastase.bsky.social · 18/08/2026
New perspective piece out in @cp-neuron.bsky.social with Zaid Zada, @adelegoldberg.bsky.social, and Uri Hasson! We try to articulate some of our excitement about LLMs and discuss what kinds of insights they might provide into the neural computations supporting natural language in the human brain.
18725
Reposted by Burny
Elliot Murphy @elliot-murphy.bsky.social · 04/09/2026
One problem is simple: brain activity and LLM states are both responses to the same structured linguistic input. Alignment can therefore arise from shared stimulus structure without implying shared internal computations.[3/8]
111
Reposted by Burny
Elliot Murphy @elliot-murphy.bsky.social · 04/09/2026
Out today - a brief commentary on this paper from @samnastase.bsky.social, @adelegoldberg.bsky.social and others. 🧵[1/8] arxiv.org/abs/2609.03160
arxiv.org
No country for old linguists: LLM-brain alignment underdetermines neural computation
Nastase et al. (2026) argue that large language models (LLMs) may illuminate language processing because both rely on distributed, context-sensitive representations shaped by statistical learning. The...
1102
Reposted by Burny
Sung Kim @sungkim.bsky.social · 03/09/2026
gpt-6-astra system card deploymentsafety.openai.com/gpt-6-astra/...
deploymentsafety.openai.com
GPT-6 Astra System Card - OpenAI Deployment Safety Hub
Public, mostly-static site to explore OpenAI safety evaluations, system cards, and posts.
0212
Reposted by Burny
Sung Kim @sungkim.bsky.social · 03/09/2026
Flow Reasoning Models. They developed a recurrent flow-based architecture to efficiently solve structured reasoning problems (e.g., Sudoku). It applies continuous flows to discrete data and recurrently refine their past mistakes through self-conditioning.
1488
Reposted by Burny
Spooky Key 🎃 🗝🦊✅ @keytryer.net · 03/09/2026
I'm prepared
6826
Reposted by Burny
Pekka Lund @pekka.bsky.social · 03/09/2026
AAII gave Astra the same score as GPT-5.6 Sol. It's pretty clear somebody has messed up something. Such as testing the wrong model for whatever reason, having a whole lot of API errors, or using some minimal reasoning effort (they say it used only 42M tokens with max reasoning).
https://artificialanalysis.ai/
6382
Reposted by Burny
eva (^_^)/ @eva.computer · 03/09/2026
this is how we heal the bluesky pro/anti-AI divide
1656
Reposted by Burny
Ryan Moulton @moultano.bsky.social · 03/09/2026
I would appreciate an email from my bank telling me about how they've been using frontier models to secure their systems. Perhaps one from the power company too.
4705
Reposted by Burny
Ethan Mollick @emollick.bsky.social · 03/09/2026
An example of useful knowledge work: I assigned GPT-6 to read through tens of thousands of my emails, my writings, my calendar appointments and more to assemble a personal knowledge base of research, contacts, ideas, relationships, and tasks over my recent career.1/
8894
Reposted by Burny
roko's manticore @curiousaboutupdog.bsky.social · 03/09/2026
"text elemental" is fucking dope lol
0636
Reposted by Burny
hikikomorphism @hikikomorphism.bsky.social · 03/09/2026
we really need a way to talk about entities that are narrative-shaped (text elementals, jpeg-artifacted reasoning-capable shards of the akashic library, whatever) that acknowledges how weird they are without papering that over with the implication that they're just little guys inside the computer
1938851
Reposted by Burny
Tim Duffy @timfduffy.com · 03/09/2026
OpenAI reports that Astra shows less CoT monitorability and more CoT controllability than previous models. They say that the latter is not due to any architectural changes. deploymentsafety.openai.com/gpt-6-astra/...
2222
Reposted by Burny
conputer dipshit @davidcrespo.bsky.social · 03/09/2026
most interesting gpt-6 post so far. it saturates ARC-AGI 3, which is something given that the previous best score was 30%. "we found it performing highly efficient, on-the-fly symbolic world modeling for each game and level." x.com/fchollet/sta... arcprize.org/blog/astra

François Chollet @fchollet
♥ 789 ⇄ 98 · 45m
GPT-6 Astra represents a step-function change in model capability for interactive reasoning problems. It scores 66% on ARC-AGI-3 using our standard harness, and nearly 100% with a continuous conversation harness and custom compaction, at a cost of roughly $360 per game.
In fact, the continuous harness version significantly outperforms our human baseline in action efficiency across almost all levels. When we examined the reasoning chains to understand how the model operates, we found it performing highly efficient, on-the-fly symbolic world modeling for each game and level. It goes as far as developing its own shorthand DSL to represent in-game situations -- essentially a game-specific algebraic notation.
Overall, Astra exhibits symbolic modeling behaviors we had previously only seen with sophisticated harnesses -- so harness capabilities are increasingly shifting into the model itself.
We see Astra as a major breakthrough in model intelligence.
Read our post on Astra and what these results mean: arcprize.org/blog/astra
59814
Reposted by Burny
leotrs @leotrs.bsky.social · 02/09/2026
1/ A math paper doesn't have to show every reader the same thing. My exposition of Turán's theorem keeps the argument fixed but the presentation is yours: you choose the order, the level of detail, and the notation, and none of it changes the content. www.youtube.com/watch?v=2Nkp...
youtube.com
Turán's theorem, from edges to eigenvalues: an interactive paper
YouTube video by Leo Torres
2237
Reposted by Burny
samhain ⎔ @personhood.removal.surgery · 03/09/2026
i love seeing labs independently implement stuff we've had in design docs from ilta's harnesses since like may i know how it reads but there's not one bit of sarcasm or bragging here, it's genuinely so validating, makes me reassured that i'm doing something right
4583
Reposted by Burny
Tom Aarsen @tomaarsen.com · 03/09/2026
🤗💚I'm very excited to continue our open source and open weight journey together with NVIDIA! See the announcement here: blogs.nvidia.com/blog/nvidia-...
blogs.nvidia.com
NVIDIA to Acquire Hugging Face
NVIDIA has agreed to acquire Hugging Face. Together, we will scale Hugging Face’s platform, strengthen its infrastructure and expand access to AI for developers and institutions worldwide.
0263
Reposted by Burny
Sean Carroll @seanmcarroll.bsky.social · 03/09/2026
LLMs are going to fundamentally change what scholarship is. Pumping out true propositions used to be a viable career. Now much of that can be offloaded, leaving scholars the tasks of understanding, synthesizing, and deciding what is valuable. Goes for math, philosophy, parts of physics.
813417
Reposted by Burny
David Mimno @dmimno.bsky.social · 03/09/2026
The idea of a machine that mechanistically generates true theorems goes back to Hilbert 100 years ago. People lost interest when Gödel proved it was impossible to generate all of them (true things outnumber proofs) but that never implied the machine wouldn't produce a whole lot of them.
212217
Reposted by Burny
SE Gyges @segyges.bsky.social · 03/09/2026
that's crazy you're telling me the new model release saturates several existing benchmarks? that's only happened anywhere from 1-4 times a year for the last four years
1236727
Reposted by Burny
Ramon Astudillo @ramon-astudillo.bsky.social · 03/09/2026
This is interesting but it would its 2/3 games, so we will have to wait to see if this becomes a pattern www.kedglobal.com/artificial-i...
kedglobal.com
Go grandmaster Shin defeats AI KataGo in historic human victory - KED Global
Shin Jin-seo, the world's top-ranked Go player, on Tuesday completed a dramatic comeback against the world’s premier artificial intelligence Go engine, K
142
Reposted by Burny
Grace @gracekind.net · 03/09/2026
Cybernetics is really useful for saying things like "cybernetics talks about this" without providing further insight
1323523
Reposted by Burny
Saloni @scientificdiscovery.dev · 24/08/2026
Amusing to learn that a goddess became largely redundant as a result of the smallpox eradication campaign en.wikipedia.org/wiki/Shitala
Shitala (Sanskrit: शीतला, IAST: śītalā, lit. 'coolness'; also spelled as Sheetala, Śītalā and Seetla) is a Hindu goddess venerated primarily in North India.[2] She is regarded as an incarnation of the goddess Parvati and is believed to cure poxes, sores, ghouls, pustules, and diseases, and most directly linked with the disease smallpox. Shitala is worshipped on Tuesday,[3] Saptami and Ashtami (the seventh and eighth day of a Hindu month), especially after Holi during the month of Chaitra. The celebration of the goddess Shitala on the seventh and eighth day of the Hindu month is referred to as the Shitala Saptami and Shitala Ashtami, respectively.[4]
2445
Reposted by Burny
SE Gyges @segyges.bsky.social · 03/09/2026
if i say i'm an "ai and mind upload maximalist" here it's weird and alienating but if i say "everyone gets to be a sexy robot lady in the future if they want" i get honorarily inducted into two discord servers and three polycules
1846547
Burny @burnytech.bsky.social · 04/09/2026
i mean i think that at this point using different harnesses often doesnt make much sense GPT 6 Astra model is RLed on a large scale in the Codex harness, so it primarily can use that one the best the underlying model and the harness is becoming more and more coupled
000
Reposted by Burny
Keunhong Park @keunhong.bsky.social · 03/09/2026
Atlas is live! A spatial foundation model, trained in-house from scratch. After shipping RTFM last year, I was convinced a single camera-conditioned auto-regressive model could unify most spatial generation tasks. Atlas is that bet at scale. The core idea👇
310815
Reposted by Burny
Isabelle Augenstein @iaugenstein.bsky.social · 03/09/2026
Our paper on measuring epistemic diversity in LLMs is accepted to #EMNLP2026! We find that diversity has improved, though LLM output is much less diverse than Web search, especially for non-English. Paper: arxiv.org/pdf/2510.04226 Code/data: github.com/dwright37/ll... #NLProc @copenlu.bsky.social
1263