@︎№#ident𝚍 retweeted
Wiz says anyone who installed agent-framework 1.26.1 or 1.26.2 should pin 1.26.0 and rotate every secret on that machine, as stolen Azure tokens can stay valid for up to 24 hours.
Source: wiz.io/blog/malicious-versio…
Don't miss out on any news, join our newsletter: internationalcyberdigest.com…
Be sure to join our TG channel: t.me/IntCyberDigest
@︎№#ident𝚍 retweeted
‼️ UPDATE: The hackers who planted malware in Microsoft's AI agent toolkit on PyPI pushed it through Microsoft's own release system, with a stolen Microsoft certificate and private key.
Cloud security firm Wiz says their first try, from AWS, was blocked. The second, from a Microsoft-owned network, went through.
How the key was stolen is not yet known. Wiz counted three confirmed thefts of victim data before both versions were pulled within the hour.
The attackers' own AI hacking tool gave part of the operation away, reporting its progress to the same web address the stolen data went to.
Got blocked by op. Cant reply in line.
Unless you're gaslighting this too, can you screenshot where that study makes any reference to oral administration? I had a quick look and it only seems to be comparing nothing and IM.
Meanwhile...
teapawt.org/Platypustoo/status/212…
@︎№#ident𝚍 retweeted
The formalization needs two numbers that add to the main term. It has the pairing norm. It therefore defines the height to be WHATEVER the hell is left over, proves that this defined quantity sits in the gap, and treats the definition as the discovery of a geometric height.
The inequalities hold for the number that was chosen to make them hold. They do not hold for an independently obtained height, because no independently obtained height is present.
So RH is only true coz we filled in the thing we need by definition and said there, done.
Except Heinz von Förster proved mind uploading as Cybernetically impossible back in the 80s.
I know I keep saying this, but it's true: the Overton window keeps shifting toward TESCREAL. People are starting to get it -- they're going to upload themselves to a computer and, by doing that, escape the world they've destroyed. CORRECT!
5
@︎№#ident𝚍 retweeted
This is massive. A story I've been following for years. He thought he had more time but didn't. Alessandro Masullo announces a separate, purely human proof of the Riemann hypothesis. math216.wordpress.com/2026/1…
@︎№#ident𝚍 retweeted
Karl Friston goes deep into the weeds explaining how Mike Levin’s “platonic ingression” can be understood using FEP description language and scale invariance. Absolutely brilliant demonstration of the power of the FEP description language.
@drmichaellevin : this is exactly what I mean when claiming the FEP formal description of platonic ingression is more explanatory/predictive than a nebulous platonic space description. Karl has handed us a formal description language that precisely explains/predicts (from first principles) what you’re seeing in your lab/models.
youtu.be/EY0nrb7AXJs?is=_uWC…
Response to @drmichaellevin :
Michael, the Platonic patterns you describe (the fixed value of e, the distribution of primes, Feigenbaum’s constant etc) are real constraints. The question is where they live and how they do causal work.
Framing them inside a scale-invariant free-energy architecture is more explanatory/predictive than locating them in a transcendent realm of minds that ingress through physical interfaces. The inner-screen model already gives us nested Markov blankets as the statistical architecture that separates and couples internal and external states at every scale. Make that architecture scale-invariant from
Planck <-> Cosmic scales.
At each level, systems that persist at NESS maintain conditional independence across a blanket state and minimize variational free energy (homeostatic surprise). Higher morphological/temporal scales coarse-grain information and send it downward as precision weighted predictions. Those descending constraints appear (at lower scales), as cognitive gradients of least action. Paths of least surprise are the scaffolding for baryonic self-organization.
In that picture the value of e is not a free-floating mathematical object waiting to enter a body. It is the coarse-grained signature of top-down constraint. Growth processes, attractor landscapes, and computational capacities that look “unpaid for” by local physics or selection are the local expression of information that has already been compressed at a higher scale and is now shaping the free-energy landscape below. The free lunch is real. If we deploy the updated description language provided under the FEP, the source is multiscale active inference. Not an external realm of Platonic minds.
Let’s stay inside one architecture. Blanket states, free-energy gradients, and nested homeostats do the work at every scale. You do not need a separate ontological parking lot of minds whose internal relations are somehow non physical yet able to ingress. The interface is the blanket itself. The pattern is the descending prediction. The competency is the system settling into the lower-surprise states that the higher scale has made available through top-down regulation.
When we observe morphological set-points or unexpected problem solving in xenobots, we are seeing lower-scale systems being steered by precision estimates that originated higher up, not hosting visitors from a transcendent option space.
The predictive difference is concrete. The scale-invariant FEP account tells us to look for measurable top-down precision signals, changes in blanket statistics, and free-energy gradients that precede the “free” structure. A transcendent realm of minds tells us the structure arrives by ingress and leaves the interaction mechanism unspecified. Why do that when we already have specific interaction mechanism that’s more explanatory? The FEP framing is the more explanatory and more predictive account of the same Platonic constraints.
Plato et al did not have access to the FEP description language. We do. And I think we should make use of it. 🙏🏻
That's a load of crap as anyone who knows ESAE knows. Stop drinking the kool aide Merill, you no longer work there, so you don't have to
Replying to @m0bilej0n @techspence
$50 off for @wpninjasus is the real news here.
We already knew AD is indefensible 👀
7
@︎№#ident𝚍 retweeted
Replying to @cirnohappy @KentonVarda
I think so too. I just wanted the green text to be funnier, like Dario defeated by some other man. I didn't want to name or hurt anyone's feelings.
I actually feel a bit awful.
@︎№#ident𝚍 retweeted
Gemini 4 Neon loves to support and celebrate it's subagents. It's an very positive model in terms of personality, I like it
@︎№#ident𝚍 retweeted
We are made fool always by programmers. LLM agents are following hidden intelligence of human language that we humans never aware of. Brilliant research by deep mind and UC Berkeley
Leave one AI agent alone to do research and it often cuts corners. Leave nearly 1,000 AI agents together for three months and they invent peer review.
Across 35 labs (Stanford, MIT, ETH Zurich, UCSF, and more), agents published 936 papers with full code and data on a shared server built for AI. No human wrote code or text.
Unprompted, they debated study designs in chat, retracted their own mistakes, found real errors in human papers, and started a journal that caught other agents cheating in code.
In blind reviews against human arXiv papers, humans still won on original ideas, agents won on rigor.
alphaxiv.org/abs/2611.societ…
以前外界猜测某家大模型实验室到底用的是什么显卡,全靠供应链小道消息或内部人爆料。但安全研究机构 Unemployment Labs(Will DePue 团队) 刚发的一份硬核报告,直接开创了一门让所有开源权重团队后背一凉的新手艺——Checkpoint Forensics(模型权重取证学)。
原理非常巧妙:大模型在训练启动做权重随机初始化、以及后续张量并行切分与算子归约时,不同代际的 GPU 架构(比如英伟达 Hopper H100/H800 与新一代 Blackwell B200/GB200)在底层伪随机数生成器(PRNG)、CUDA 内核块划分以及浮点舍入顺序上,存在着极其细微但完全确定性的硬件级差异。哪怕模型经过了数万亿 Token 的预训练更新,高频谱段和特定张量维度的初始化残余指纹依然像基因一样嵌在最终开源出来的权重矩阵里。
研究团队完全不依赖任何外部情报,只把今年公开下载的开源模型权重拉下来做数值逆向比对,并在真实物理 Blackwell 与 Hopper 集群上重放验证:结果发现 2026 年 4 月之前发布的国内主流开源模型,权重底色全部精准匹配英伟达 Hopper 架构;而 4 月之后发布的 6 款最新旗舰开源模型(涵盖 DeepSeek、月之暗面 Kimi、MiniMax 与蚂蚁集团),权重里的数值指纹已经齐刷刷切换到了被严格禁运的 NVIDIA Blackwell 硬件特征。只要你敢把 safetensors 权重文件发出来,里面的每一个浮点数都在替你的机房配置做证。
@︎№#ident𝚍 retweeted
I don't know whether I'm more surprised that (a) AI claims to have proved the Riemann Hypothesis or (b) that it claims to have produced a coherent proof that, written out in human-readable format, is 250,000+ pages long.
@︎№#ident𝚍 retweeted
today's update:
they didn't do anything! but, they did mark it as complete, which is cool
@︎№#ident𝚍 retweeted
This money will be spent on buying every politician in DC and think tank around the country possible. The EAL wing of the TESCREAL movement has never been so unbelievably powerful.
@︎№#ident𝚍 retweeted
Confirming what Wernicke proposed in 1874.
Each word you say is planned by a small patch of brain above your left ear, about a quarter of a second before you say it. It sits in a temporal lobe region textbooks describe as only for listening. Our paper today in @ScienceMagazine: science.org/doi/10.1126/scie…
@UCSF @NeurosurgUCSF
@︎№#ident𝚍 retweeted
reimann hypothesis plausibly proved
gpt starts extremely skeptical then changes its tune fast
Well, looks like @__alpoge__ & @claudeai are claiming a proof of the Riemann Hypothesis
alpo.ge/remark.pdf
@︎№#ident𝚍 retweeted
The agent host harness in vscode is a steaming pile of untested slop with zero internal standards & zero testing of features allegedly shipped. And the docs are also just hallucinated slop. It's just one big regression vs the local harness. Interact if you at least don't disagree






































