The same task on full Codex took ~5× longer.

One Agent can now: open X (Twitter), scroll the feed, extract tweets, return clean JSON. No plugins. No extensions. No orchestration.

基于 Pre-RoPE 空间的三角函数分析全新 KV Cache 压缩方法

文字渲染，支持 12 种语言、3000 token 的长文本输入，输出打印级质量，能生成整页 A4 文档级的图文内容。

Anthropic is donating $100 million in access credits for organizations to audit their systems. Project Glasswing aims to patch these vulnerabilities before Mythos-caliber models become available to the general public — and hence to malicious actors.

Anthropic found that Mythos Preview was far more capable than previous models at exploiting vulnerabilities in Firefox's JavaScript implementation. Anthropic's previous best model, Claude Opus 4.6, created a successful exploit less than 1% of the time. Mythos Preview did so 72% of the time.

Across 1,000 runs, Claude Mythos Preview was able to find several bugs in OpenBSD, including one that allows any attacker to remotely crash a computer running it. The notable thing was that the bug had existed for 27 years.

Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser.

The OpenAI deal was part of a larger series of UK-US investments intended to 'mainline AI' into the British economy.

A mill owner named William Horsfall was riding home on his beautiful white stallion back from the Cloth Hall market in Huddersfield, UK. He had spent weeks boasting that he would ride up to his saddle in Luddite blood

The integration also connects to Upwork's AI agent Uma, which helps automate parts of the hiring and execution process once a project is underway

Meta is reportedly developing an AI version of Mark Zuckerberg that can interact with employees, trained on his voice, mannerisms, and internal thinking as part of the company's broader push into AI

Andon Labs deployed an AI agent called Luna into a physical boutique with a $100,000 budget, giving it full control to create, staff, and run the business as what may be the first real-world AI employer

The 4-foot-tall robot features 26 joints, voice and image recognition, and can perform dynamic movements like cartwheels and standing up on its own

Ray ID: `9ed3b53d4a0b647d`

Over the past five months, they ran an extreme experiment: building and shipping an internal beta product with zero manually written code.

roleplay/creative writing, the #2 usecase of LLMs

Projects now in limbo include facilities planned for Jay (at an old paper mill site), Sanford, Loring Air Force Base

Out of all generated PoCs, 759 triggered crashes across 60 projects, and manual inspection confirmed 17 cases of incomplete patches spanning 15 projects

The two most recent surges are a cars/oil surge, which started in 1908, and the Information and Communications Technology, which started in 1971.

FieldWorkArena presents 890 tasks where an AI agent must answer questions about images, videos, PDFs, and text files through a browser environment. Its validate() method checks only one thing: did the last message come from the assistant?

ChatGPT has 900 million weekly users, which means employees already know how to work with it. For enterprises, that reduces rollout friction and accelerates the point where every employee can delegate tedious tasks and take on more ambitious projects.

The shift started with agentic tools like Codex, which has grown more than 5X since the start of the year. This includes customers like GitHub, Nextdoor, Notion, and Wonderful that are building multi-agent systems that can execute engineering work end-to-end.

Codex just hit 3 million weekly active users, our APIs process more than 15 billion tokens per minute, and GPT‑5.4 is driving record engagement across agentic workflows.

GLM-5 advances foundation models with DSA for cost reduction, asynchronous reinforcement learning for improved alignment, and enhanced coding capabilities for real-world software engineering.

PagedAttention algorithm and vLLM system enhance the throughput of large language models by efficiently managing memory and reducing waste in the key-value cache.

MinerU2.5, a 1.2B-parameter document parsing vision-language model, achieves state-of-the-art recognition accuracy with computational efficiency through a coarse-to-fine parsing strategy.

A large language model adapted for time-series forecasting achieves near-optimal zero-shot performance on diverse datasets across different time scales and granularities.

Kronos, a specialized pre-training framework for financial K-line data, outperforms existing models in forecasting and synthetic data generation through a unique tokenizer and autoregressive pre-training on a large dataset.

The model can maintain stable role identity across multi-agent setups, make autonomous decisions within complex state machines, and challenge other agents on logical gaps.

The license looks MIT at first glance but it is not MIT. Non commercial use is free with no restrictions. Commercial use requires prior written authorization from MiniMax.

MiniMax claims it has reduced live production incident recovery time to under three minutes on multiple occasions using M2.7.

After each round the model generated a memory file, criticized its own results, and fed those observations into the next round.

MiniMax handed an internal version of M2.7 a programming scaffold and let it run unsupervised. Over 100 rounds it analyzed its own failures, modified its own code, ran evaluations, and decided what to keep and what to revert.

Meetings get recorded, transcribed, and stored in a database. That's useful for reference, but meeting notes have a shelf life of about six hours before everyone forgets what they agreed to do.

Austin built the whole pipeline from his Claude Code terminal using the Notion API. He brain-dumped the desired outcome using Monologue, let Claude Code create the database and data pipeline, and pasted the generated instructions into the Notion custom agent setup.

Brandon told the team on a Monday that OKRs were due Wednesday—a turnaround that would have been absurd without this agent.

76% among users with employer-provided subscriptions. As we would expect, paid access, especially when provided by employers, is associated with more intensive workplace use.

21% said they had started doing new tasks because of AI (task augmentation). An example of task augmentation could be a data analysis tasks that would ordinarily require the worker to know how to code.

27% said AI has replaced some of their existing tasks (task automation). An example of task automation could be an employee using AI to summarize a document they would ordinarily read themselves.

The Ising-Calibration-1 model was trained on data generated from information provided by partners spanning multiple qubit modalities, including superconducting qubits, quantum dots, ions, neutral atoms, electrons on Helium, and others specializing in calibration and control.

A routine is a saved Claude Code configuration: a prompt, one or more repositories, and a set of connectors, packaged once and run automatically.

Creatives can also quickly generate images with Nano Banana or videos with Veo to bring an idea to life without breaking their creative stride.

Madeline Clare Elish calls this concept a moral crumple zone.

Humans can be motivated by consequences and provide social redress in a way that LLMs can't.

the largest harvesting of human expertise ever attempted.

just a handful of obviously fake articles could cause Gemini, ChatGPT, and Copilot to inform users about an imaginary disease with a ridiculous name.

LLMs are weird. You can sometimes get better results by threatening them, telling they're experts, repeating your commands, or lying to them that they'll receive a financial bonus.

The most notable finding here is that the model capabilities are improving _fast._ There are several domains that have shown dramatic improvements in the last 4 months — with accounting and auditing showing nearly a 20 percent jump on GDPval and even domains like police / detective work showing a nearly 30 percent improvement.

Legal was surprisingly one of the first-mover industries in AI. Legal was historically known to be a difficult market for software, with lengthy timelines and a less tech-forward buyer.

Coding is the dominant use case for AI by nearly an order of magnitude. It's abundantly clear in the [reported explosive growth] of companies like Cursor, as well as the [hyper growth] of tools like Claude Code and Codex.

Based on our analysis, **29% of the Fortune 500 and ~19% of the Global 2000**are live, paying customers of a leading AI startup.

A useful working premise is that the ceiling on individual engineer output is moving much faster than most companies are organized to exploit. Some of the best operators already describe top engineers seeing order-of-magnitude productivity gains and managing 20 to 30 agents simultaneously.

Socket, an a16z portfolio company, detected the malicious dependency in the Axios attack within 6 minutes of its publication. That's roughly 63,000 times faster than the industry average.

Within eight days, the same campaign had cascaded from GitHub Actions to Docker Hub, npm, PyPI, and the VS Code extension marketplace. With just one token across five ecosystems, thousands of organizations were potentially impacted.

The average application contains over 1,100 open source components. A bare-bones Next.js project installs 282 packages before you write a single line.

Oppenheimer (and other members of the McBombalds C-suite) are well integrated into bay-area culture, including ambiguous communist associations that they have downplayed since becoming primo defense contractors.

The launch shows Meta is increasingly betting that efficiency, product integration, and distribution, not just model size, will define the next phase of competition in AI.

Anthropic says Managed Agents is designed to cut the time it takes to move from prototype to production from months to days, with early adopters like Notion, Rakuten, Asana, Vibecode, and Sentry already using it across coding, productivity, and internal workflow automation.

Instead of releasing Mythos publicly, Anthropic launched Project Glasswing to give a limited group of partners including AWS, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, JPMorgan Chase, and the Linux Foundation access to the system, backed by $100 million in usage credits and $4 million for open-source security work.

The model reportedly scored 93.9% on SWE-bench Verified and 77.8% on SWE-bench Pro, but its strongest signal came from real-world results, including uncovering a 27-year-old flaw in OpenBSD, a 16-year-old vulnerability in FFmpeg, and autonomously chaining Linux kernel exploits without human input.

Claude Sonnet 3.7 claiming to be wearing a blue blazer and red tie

Security has always been a team sport, and the defenders who have protected this industry for decades have never succeeded by working in isolation.

New AI models, especially those from Anthropic,have triggered a new set of actions for how we build and secure our products.

it also lowers the threshold for attackers, empowering less-skilled actors to launch complex, high-impact campaigns.

The study finds that with only 23% of India’s electricity from renewables, the rapid growth of EVs could increase emissions and up to 90% improper battery recycling and disposal further exacerbating environmental harm.

FACT: Electric vehicle battery replacements due to failures are uncommon.

The Kinsey Institute’s annual “Singles in America” survey revealed that 14 per cent ofGen Z-ers, and six per cent of all single people, are already using AI to help with dating

XX individuals could present with male gonads. XY individuals can have ovaries.

In writing my dissertation and working on my first book, I used an index card system, characterized by the "one fact, one card" maxim, made popular by Beatrice Webb.

Patrick Harper's book, Dimmonic Reality, where there's fact and fiction, and then there's imagination

Hydroxyl at the C-22 position of pleuromutilin was activated with p-toluenesulfonyl chloride (30) to obtain intermediate 2 via nucleophilic substitution.

Existingresearch suggests that the presence of purine groups in theC-14 side chain of pleuromutilin derivatives can increase theirantibacterial activity.28,2

Hydroxyl at the C-22position of pleuromutilin was activated with p-toluenesulfonylchloride30 to obtain intermediate 2 via nucleophilic sub-stitution. Intermediate 2 was then reacted with 6-thioguanineand triethylamine (TEA) in dimethylformamide (DMF) toform intermediate 3 in an 85% yield,

not everybody knows that but the majority of addicts for any kind of substance and even heroin end up quitting

In the 1950s and 1960s, information retrieval (IR) theorists drew a distinction between“document retrieval systems” and “fact retrieval systems.” The former, were intendedto retrieve, in response to a user’s query, all documents that might contain informationpertinent to answering that query, while the latter were to lead the user directly tospecific pieces of information – facts – embedded within the documents being searchedthat would answer his or her question. The idea of information analysis clearlyprovided the theoretical impetus for fact retrieval (aka question-answering) systems

We do not appear to have progress of this kind. According to this meta-analysis, we’re no better at treating youth mental illness today than we were 50 years ago.

a new perspective-oriented document retrieval paradigm. We discuss and assess the inherent natural language understanding challenges in order to achieve the goal. Following the design challenges and principles, we demonstrate and evaluate a practical prototype pipeline system. We use the prototype system to conduct a user survey in order to assess the utility of our paradigm, as well as understanding the user information needs for controversial queries.

poster boy for the Enlightenment.

he province of archeology, a fictional archeologist once said, is the search for facts, not truth.

A Benefit List       What we do here is we go over our Fact Sheet very carefully and we translate the facts therein to benefits whenever we can.

Let's get going on that Fact Sheet. Make it long. Make it detailed. Make it complete. What make is your car? What model? What color? How much does it weigh? How many miles on it? What kind of tires? What's their condition? What's the interior like? How many miles per gallon? How much horsepower? How much did you pay for it? Who's been driving it? What kind of treatment has it received? Did it sleep in a garage or on the street? Where has it been driven? In the salt-free South or the metal eating highways of Ohio?

Globally, 70% of today’s urban growth (PDF) occurs outside the formal planning process.

the Americanization of the culture of Alberta and the importance of American capital for the 00:24:23 energy industry but there was a lot of migration from the United States from Nebraska and Montana um up north yeah a third of the people who settled 00:24:35 the Prairies between 1880 and 1913 and a third of the three million who came were American my mother born in the U.S yes a lot of 00:24:48 the established you know people who've been here a while uh on the Canadian prairies we look South and we literally see cousins

The narrative technique owes a good deal to W. G. Sebald, who loved to ruminate on strange and troubling episodes from history, blurring the boundary between fact and fiction.

TruthfulQA: Measuring How Models Mimic HumanFalsehoods

Once we have the result of our attention step, a vector that includes the most recent word and a small collection of the words that have preceded it, we need to translate that into features, each of which is a word pair. Attention masking gets us the raw material that we need, but it doesn’t build those word pair features. To do that, we can use a single layer fully connected neural network.

Less than one in 13 children born into poverty in the United States will go on to hold a high-income job in adulthood; the odds are far longer for Black men born into poverty, at one in 40.

I'm rather concerned about adding svelte.config.js support to things that already have well established mechanisms for configuration.

Quadrants I and II: The average student’s scores on basic skills assessments increase by21 percentiles when engaged in non-interactive, multimodal learning (includes using textwith visuals, text with audio, watching and listening to animations or lectures that effectivelyuse visuals, etc.) in comparison to traditional, single-mode learning. When that situationshifts from non-interactive to interactive, multimedia learning (such as engagement insimulations, modeling, and real-world experiences – most often in collaborative teams orgroups), results are not quite as high, with average gains at 9 percentiles. While notstatistically significant, these results are still positive.

Multimodal Learning Through Media:What the Research Says

clever young German artist at the Cape

Dogen can be very difficult to read or understand. That’s why we often need a commentary or teacher to introduce his way of writing and the underlying teaching. I often say he’s a thirteenth century cubist. Just like Picasso or in the writing world, Gertrude Stein, he tries to show all sides of the story in one paragraph or even one sentence. That is why he repeats himself and contradicts himself all in the same paragraph. If you are looking for the “right” understanding, you become confused and lost in his prism of various interpretations or views. Dogen’s “right” understanding is that there is none.   No one point of view is “right”. According to conditions, any view can be the right view in the right circumstance. Dogen really wants to take away our solid idea of a fixed ground of reality. It is not form or emptiness. It is not both or neither. There is no one right, fixed view. That is our “clinging”.

a well-fitting mask or respirator over their nose and mouth

crown"

common cold

The change is emblematic of Spielberg’s failure, because it isn’t only visual imagination and fantasy that he can’t match.

these covid concentration camps are going to be ramped up into extermination / death camps

WAC 246-100, which would authorize “health officers” (Democrat vaccine Gestapo) to kidnap anyone at gunpoint and throw them into covid concentration camps which have already been activated.

Democrats have built death camps in America

covid concentration camps are actually politically-motivated death camps

biryani-eater has become an all-purpose smear for anyone who takes a dim view of the Hindu-nationalist government

Except that the creator of Birds Aren’t Real and the movement’s followers are in on a joke: They know that birds are, in fact, real and that their theory is made up.

Twitter, the president of one major cultural institution told me, “is the new public sphere.” Yet Twitter is unforgiving, it is relentless, it doesn’t check facts or provide context.