stonebig..

[stonebig.. @stonebigdotdot]

so obvious now

[Haider. @haider1]

"the IT department is going to become the HR department of your digital workforce"

Nvidia CEO, jensen huang predicts:

IT departments will evolve into HR for digital workforces which are responsible for onboarding, fine-tuning, and continuously improving the AI system

[David Brochart @davidbrochart]

Jupyverse, the FastAPI-based Jupyter server, now has a libghostty-based terminal emulator 🚀

The original goal is to allow reconnecting clients to recover the current terminal state, but who knows what it will unlock in the future!

[Phoronix @phoronix]

LZ4 1.11 Introduces Multi-Threaded Compression For liblz4, Other Performance Gains

https://t.co/nPUWoXVGzj

[Phoronix @phoronix]

LXQt 2.5 Released With Enhanced QTerminal, Removing The Final LXDE Dependency

https://t.co/kekSdQWTiy

[Khuyen Tran @KhuyenTran16]

Polars 2.0 mainly changed how existing code behaves. Polars 2.1 brings a more exciting set of new features.

Here are three I found most useful:

• Run group-bys faster with less memory
• Use more warehouse-style SQL functions
• Encrypt selected columns in Parquet files

Polars 2.1 also adds support for conditional joins, approximate quantiles, and masked NumPy arrays.

This article demonstrates each feature with concrete examples, outputs, and diagrams.

🚀 Full article: https://t.co/7KesxOhdh8

#DataEngineering #Python #DataScience #BigData

[Yann LeCun @ylecun]

LLMs accumulate a lot of declarative knowledge.
That's how they can answer a large number of questions.
That's also why they require so many parameters and hence so much memory.

But one can replace the accumulation of knowledge by *understanding*
That what JEPA world models do.
That's one reason they can be useful with considerably less memory and training data than LLMs.

This paper merely shows that the underlying similarity structure of the data, whether text or image (e.g. co-occurence of words in text or objects in images) uniquely determines the shape of the representation point cloud, when the encoder is trained trained with a common anti-collapse criterion like SIGReg or VCReg (not by reconstruction).

[Simon Willison @simonw]

There's a buzzy new paper about GPT-6 Astra being able to directly interpret model weights (prompts full of floating point numbers) - I tried it with Qwen 3.8 27B on my laptop and found that, after 45+ minutes, that model could achieve the same thing https://t.co/6O7WltGP7K

[Gregory Kennedy @gregorykennedy]

This is the single most insightful thing I have ever heard about the changing landscape of modern business. Haters are gonna hate, but there is a reason he is the most successful venture capitalist of all time. And no, it's not just luck.

Respect. 👊

[alphaXiv @askalphaxiv]

Yann LeCun and Randall Balestriero just Proved the Platonic Representation Hypothesis, and found a better way to achieve it.

The Platonic Representation Hypothesis suggests that AI models trained on different modalities, like images and text, eventually learn similar representations of the world. But this paper shows that how models are trained matters more than simply scaling them up.

So they mathematically show that, under certain conditions, two JEPAs trained completely separately can learn essentially the same underlying representation of the world. One learns from different images of the same scene, while the other learns from different descriptions of that scene, without ever being trained on image-text pairs.

Their approach achieves the same image-text alignment with 42x fewer training samples than next-token prediction. A 109M-parameter text JEPA even outperforms LLMs up to 72B parameters on cross-modal alignment benchmarks.

https://t.co/3j1enq9qF7

[stonebig.. @stonebigdotdot]

Pure objectivity swamps the world

[Satya Nadella @satyanadella]

1/ We announced our biggest update to Copilot to date, bringing together Chat, Cowork, Autopilot, and Code into one unified experience:

· Daily users grew nearly 50% QoQ, and more than 5X YoY.
· User satisfaction improved nearly 20% QoQ to another all-time high
· Number of customers with more than 50K Copilot seats, up 8X YoY

Great momentum with GitHub Copilot too, including Copilot Fusion, announced just this morning, which delivers frontier level quality, at up to 77% lower cost.

And this is just the beginning. The next frontier is maximizing intelligence per task while continuously optimizing for cost and performance.

This is the frontier that will matter to companies!

And we are extending our vision of Copilot as the new OS for work, bringing the business logic and capabilities of D365 directly into Copilot, helping every organization stand up their own infinite SaaS factory.

[stonebig.. @stonebigdotdot]

Nice ai generated music. There was the same feel of ending before quantum theory, then before quantum entanglement.

[Simon Coste ꙮ @__SimonCoste__]

It's 2026. Your lab throws the biggest Halloween party to celebrate the Death of Mathematics. PIs are wasted in the coffee room, Phds and postdocs dance around a bonfire from obsolete preprints, even Reviewer 2 is having fun

And then the DJ (Fields medal) throws this banger 🎉

[Simon Coste ꙮ @__SimonCoste__]

It's 2026. Your lab throws the biggest Halloween party to celebrate the Death of Mathematics. PIs are wasted in the coffee room, Phds and postdocs dance around a bonfire from obsolete preprints, even Reviewer 2 is having fun

And then the DJ (Fields medal) throws this banger 🎉

[stonebig.. @stonebigdotdot]

At this pace , Riemann hypothesis and grand unified theory will be under the Christmas tree

[Sughu @sughanthans1]

openai says its ai just solved three famous open problems at once: one in math, one in computer science and one in physics.

the math one has been open since 1991. the computer science one is an assumption hundreds of papers were built on, and openai says it's wrong. the physics one touches a millennium prize problem. outside experts are now checking all three, so none of it is settled until they're done.

1. math: waves can't stay focused for long

picture ripples in a pond all rushing toward one spot. for a split second they pile up into a tall spike, then spread out again. in 1991 a mathematician named sogge guessed that this extreme focusing is always brief, and that at a random moment a wave is much calmer than at its worst.

it was proved in 2 dimensions in 2019. openai says its model proved it in every dimension, which also settles two other famous problems, the kakeya and restriction conjectures. the next day it was checked in lean, a program that only accepts a proof if every single step is correct.

2. computer science: a shortcut that "shouldn't exist"

some puzzles seem to force you to try nearly every combination, like a lock where the only way in is testing code after code. the exponential time hypothesis (eth) is the widely believed idea that for puzzles like this there's no big shortcut.

hundreds of papers in graph theory, biology, databases and more lean on it to say "this can't be done much faster." openai says its model proved eth is false, also checked in lean. that reopens how much faster a lot of those tasks could get.

the researchers who first proposed eth, impagliazzo and paturi, are leading the group checking it, with ryan williams. the proof is public for anyone to look at.

3. physics: putting the strong force on solid math

the strong force is what holds tiny particles called quarks together inside every atom, using carrier particles called gluons as the glue. physicists describe it with a theory called qcd and use it constantly, but nobody had proved it holds up as rigorous math.

openai says its model proved it does, and did the same for a unified version of the standard model, physics' main theory of particles. one piece of the work, proving that yang-mills theory exists, is part of one of the clay institute's millennium prize problems.

the physics group is led by david gross, who shared a nobel prize for work that helped establish qcd. they're holding the ai's work for now and writing it up themselves.

openai says it wants no academic credit for any of this. the experts checking it are unpaid, signed no ndas, and are free to publish whatever they find.

[Sughu @sughanthans1]

https://t.co/Wg719CoojH

[Lex Fridman @lexfridman]

This is an incredible moment in this history of mathematics, computer science, physics & science with some of the hardest problems in human history being solved by AI over a period of weeks. Depending on your perspective, it's either a crisis, a triumph, or both. I don't think anyone knows for sure. We're all trying to make sense of it. I oscillate between fear/dread and optimism/excitement on a daily basis. The rate of progress is insane, including today's announcement from OpenAI of more problems solved.

To help make sense of all of it, here's my conversation with three mathematicians about this moment. Each of them is an expert in the field of each of the biggest problems solved: Navier-Stokes, quasi-Riemann hypothesis, the unique games conjecture and complexity theory more generally.

Ankur Moitra is a mathematician and theoretical computer scientist at MIT, visiting the Institute for Advanced Study in Princeton. Steven Brunton (@eigensteve) is a mathematician and fluid dynamics expert at UW. Alex Kontorovich (@AlexKontorovich) is a mathematician at Rutgers specializing in number theory.

[Eric S. Raymond @esrtweet]

The only way out is through.

They're calling it the Bugpocalypse. LLMs are being used to find long-hidden cracks in the software infrastructure, and it's happening at a furious rate. 3% of all compromises in computing history happened in the last week, claims one source? I believe it.

There are people saying public access to the frontier models ought to be locked out, or heavily guardrailed, over this. The problem with that idea can be summed up in one sentence: when lockpicks are outlawed, only outlaws will have lockpicks.

The lockpicks will be Chinese open-weight models, if nothing else. There's no way to stop those from becoming ubiquitous.

There are two futures in front of us. One in which only the bad guys have lockpicks, and security holes go undiscovered until they're exploited. Another in which everybody has lockpicks and is aggressively red-teaming their own software so it gets fixed before the holes are exploited.

The corpus of human code is huge but finite. The number of defects in it has to be finite too. We can fix those defects now; we have a million friendly robots to do it with.

The only way out is through.

[stonebig.. @stonebigdotdot]

@geoffreyirving Natural evolution <> preventive evolution

[AYi @AYi_AInotes]

网景创始人 、硅谷顶级投资人Marc Andreessen 在访谈里坦言，他看到了科技史上最可怕的一道认知鸿沟，

传统巨头以为 AI 只是个降本增效的插件，而新一代 AI 原生团队，正在用完全非人的组织效率掀翻整个商业版图。

当年写出网景浏览器、见证了整个互联网三十年兴衰的 Marc Andreessen，坐在 Ben Horowitz 面前直接发出了感慨，
如果 1994 年我们做 Netscape 的时候就有今天这种 AI 工具，那简直不可想象。

现在绝大多数传统巨头和现有企业，依然在用改良软件的旧思维去看待大模型。

在他们的组织逻辑里，AI 只是一个高级功能，用来给现有业务做个插件、在钉钉里加个智能助手、或者把客服团队砍掉 20% 省点成本。

他们以为自己在拥抱变革，其实只是在给老旧的蒸汽机打补丁。

但真正恐怖的事情，正在那些刚成立的 AI 原生团队里发生。

那些创始人对 AI 能力的理解，跟大厂完全不在同一个平行宇宙。

他们每周都在看着底层技术飞速进化，而且他们压根不会把自己定义成一家软件公司、硬件公司、甚至是一家 AI 公司。

以前创办一家公司，你的组织架构是按人类的协同瓶颈搭出来的：
招产品经理写文档、招工程师写代码、招法务改合同、招运营做投放。
因为人类的脑力带宽有限，所以必须靠庞大的部门和管理层级来支撑业务。

而真正的 AI 原生物种，是用机器智能作为最底层的骨架。

两个人加上一整套自动化 Agent，可以直接把从市场洞察、软件工程、全栈设计到商业销售的链路全部跑通。

他们不是在效率上快了 30%，他们是在组织形态上彻底跨越了一个物种。

大厂的盲区，从来不是买不起算力，而是他们无法理解组织本身才是最大的负债。

当现有企业还在开会讨论如何把 AI 塞进现有的业务流程时，新一代的野蛮人正在用完全超越行业界限的方式，把传统的业务流程连根拔起。

未来拉开人与人、公司与公司之间差距的，根本不是你用了哪个模型，而是你到底是在用新工具维持旧世界的体面，还是用新世界的底层逻辑，重新定义了自己能做的事情。

[Jason D. Clinton 🔸 @JasonDClinton]

as bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better. security leaders, never let a crisis go to waste: please go to your boards and ask for help right now. not for ai, for the basics. (we need to hold on for ~18 months.)

[Anthropic @AnthropicAI]

New on our Frontier Red Team blog: using Claude to formally verify software at large scale.

Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).

With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.

Read our research in full here: https://t.co/zUyx8CjBEk

[stonebig.. @stonebigdotdot]

security weather of October 20th 2026 was as bad as it could get, so far.