The functionality seamlessly supports everything from basic arithmetic to highly intricate calculations, simplifying what is traditionally a frustrating and time-consuming debugging process.

This ensures that both novice users and seasoned data analysts can maintain momentum without having to manually parse error messages or search external forums for solutions.

We also introduce an agentic streaming inference framework that supports thousand-second-scale generation while mitigating drift.

Forcing-free streaming training. Multi-stage training enabling native, efficient streaming audio-visual training at 22B scale.

They would be very happy to have a tool that does one wafer per hour and it costs them a fortune to run. They would build a fab with a thousand of those and be super happy with it.

The Japanese robotics company FANUC is itself one of the original dark factory pioneers that has operated a 'lights out' factory since 2001. In other words, the FANUC robot arms being deployed by GM and other companies to automate automotive production were themselves primarily built by other robots.

Recent events highlight how important open source is to the AI ecosystem, with more nations and enterprises recognizing the risks and costs associated with exclusively depending on closed models.

Reflection has leaned directly into that pitch as the startup, last valued at $25 billion, is trying to build American open-source AI models that can compete with frontier systems from OpenAI, Anthropic and Google.

we can finally invent new products that allow users to do things more naturally, using simple language to express their needs.

when I first experienced OpenClaw earlier this year, I had the epiphany that it isn't the models that matter, but the harnesses, loops, and context which will lead to so many new opportunities ahead.

For agent products, that may matter more than raw benchmark scores.

Fugu models surpass publicly accessible frontier models and are shoulder-to-shoulder with Fable 5 and Mythos Preview in various rigorous engineering, scientific, and reasoning benchmarks while delivering frontier capability without the risk of export controls.

Collective intelligence serves as the practical hedge against this concentration of power.

orchestration is no longer just a technical optimization; it has become a geopolitical and operational imperative.

AI may generate an insight, but people must still evaluate its significance and plausibility.

That was the moment that I felt like, okay, these models have now come to a point where they really, truly understand.

How Codex helps work continue beyond a single prompt

Our models identified a 23-year-old use-after-free in OpenBSD's kernel implementation of System V semaphores.

The completed setup took less than a day. Trail of Bits estimates that building the same lab manually would ordinarily take at least several weeks.

OCR 4 is a document-understanding model, not a decision-maker. It is not intended for medical diagnosis, legal advice or judgment

Models building their own software tools might have seemed outlandish not long ago, but it is happening. It would be unwise to rule out the same trajectory in hardware.

We are plausibly entering the early era of physical agentic AI.

it was as or more successful than both human teams while producing almost ten times less code than Team Claude.

Claude Opus 4.7—operating without human assistance—was about 20 times faster than the fastest human team at all tasks completed by our participants less than a year ago.

Claude can even automatically learn from _other_ Slack channels and data sources, if it's granted permission.

AI300 with HBC Gen 2 is designed to enable another stepwise improvement with a 54x increase over AI200

The scale is what is new. Earlier automation bolted fixed arms to a line. Humanoids move anywhere and vendors pitch them to do almost any manual job.

Hyundai talks about safety and labour shortages. The union talks about jobs and bargaining power. Both describe the same machine.

Anthropic contends that the cited breach was a narrow jailbreak, one that rival models, including OpenAI's GPT-5.5, also exhibit. According to the company, the flagged behavior amounted to asking the model to analyze a codebase and fix identified issues, which revealed a few minor, already known bugs, rather than a genuine autonomous offensive intrusion.

HappyHorse is built around a 15-billion-parameter unified self-attention Transformer that processes text, image, video, and audio tokens within a single token sequence. Unlike many competitors that stitch together separate models for video and audio

The fact that these smart glasses truly looked like ordinary glasses you wouldn't be ashamed of wearing was a simple but inspired design choice.

Unlike the ChatGPT or Claude app, Siri AI is woven right into the iPhone, so it's even more ready to go beyond answering questions and start automating more aspects of the user experience.

In the case of aircraft, most nations cooperate on things like safety standards and air traffic control.

The AI industry has reached the stage where it can't just be exciting and new anymore. It has to prove its worth.

We now appear to be entering the era of token rationing.

Previously only available as a standalone Gemini 2.5 computer use model, computer use is now integrated natively in the main Gemini Flash model.

Computer use is now a built-in tool supported in Gemini 3.5 Flash, delivering our best performance yet for agentic computer use tasks.

xAI struck a deal to give Cursor access to its compute infrastructure, foreshadowing similar, larger deals with Anthropic and Google in the future.

Those deals with Anthropic and Google have relatively favorable termination clauses for SpaceX, so if SpaceX's enterprise AI efforts take off and see high demand, it will theoretically be possible to reallocate compute from competitors directly to SpaceX and the Cursor team.

We find that GLM-5.2 shows more potential hacking behavior than GLM-5.1. This makes the verification signal easy to optimize, but fails to actually improve the fundamental capabilities of the model.

Instead, with IndexShare, the KV cache of $h_5$ includes only $kv_{1:4}$, all from the hidden states of the target model. For training, we reuse both kv cache and topk indices of the first mtp step.

To address this, we introduce an anti-hack module for both RL training and evaluation. The detection process has two stages: a rule-based filter first catches potential hacks to maximize recall, and then an LLM judge checks the intent of these flagged actions to keep precision high.

The US government's Anthropic models ban was never about an AI jailbreak

Trump may see restricting Mythos and Fable as a matter of national security. But the argument cuts both ways, and with Washington now asking if AI is too important for everyone to have access, other governments are asking whether they can afford for Washington to decide who does.

He likened the pullback of Anthropic's models to Iran's blockade of the Strait of Hormuz, with access to AI now a strategic chokepoint for which France must prepare.

Commerce dept. worries that a Fable 5 'jailbreak' could be a national security threat.

The company says it has only seen evidence of this kind of jailbreak being used to find 'minor' and 'relatively simple' software vulnerabilities

We stand by this defense in depth strategy. It reduces the risks posed by Fable, making them comparable to the risks of existing models already deployed across the industry.

We have found that other publicly-available models are able to discover them as well without requiring a bypass.

If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

The potential jailbreaks that have been disclosed to us are either entirely benign responses or are minor findings that provide no Mythos-specific uplift.

We suspect that perfect jailbreak resistance is not currently possible for any model provider.

We stand by this defense in depth strategy. It reduces the risks posed by Fable, making them comparable to the risks of existing models already deployed across the industry.

We have found that other publicly-available models are able to discover them as well without requiring a bypass.

apparent hallucinations

Once again, AI proves to be an unreliable source of information about AI.

His personal cost of capital made that possible.

At inception, cost of capital is purely personal. Founders & an idea. No business exists yet to evaluate.

The company changed course after the move received significant backlash from the AI research community.

The concern is that as more and more AI agents get deployed and begin working together, we could hit a tipping point where imagined scenarios become real.

Some researchers, including a team at Google DeepMind, have argued that artificial general intelligence could come not from a single super-smart model but from a kind of agent hive mind, where the capabilities of the whole add up to more than the sum of its parts.

When we talk again in three months, we'll be like, 'Here's 100 developers that all built 100 different applications with Oasis that surprised all of us,'

But by letting you generate a world for so long, the model also degrades significantly.

Composer 2.5 is exceptionally intelligent & up to 10x more efficient than similarly capable models

Frontier model prices keep rising for the smartest models

current agent performance is still strongly shaped by harness behavior and workflow choices, not just base-model quality

Even with extended thinking time (10,000 tokens), Python access, and the ability to run experiments, success rates remained below 2%—compared to over 90% on traditional benchmarks.

Many SWE-bench-Passing PRs Would Not Be Merged into Main

The headline result is that the best model, Opus 4.8, scores only about 13% on the hardest subset—far below the 50%+ regime common on SWE-Bench-style evals

agents often lack a dependable way to access the databases containing the information they need.

adding a deterministic retrieval layer made model choice much less important

When we have [artificial general intelligence], I don't think there will be a large number of distinct brands, said Alex Embiricos, OpenAI's head of enterprise product.

The changes underline how OpenAI's strategy is moving closer to that of Anthropic, whose focus on developing products for businesses has stoked its blistering growth.

Notion said it was disabling use of 'all Anthropic models' in its automated productivity tool.

Is there any way that these labs can squeeze pennies like Uber has squeezed the drivers over the years? Is there something squishy enough there for them to do that?

the whole tokenmaxxxing thing has become a thing, peaked, and now is seen disfavorably, within six months.

This hybrid-architecture trend with alternating attention and alternative layers is a relatively popular development this year

long-context efficiency is king as more and more LLMs get plugged into agent harnesses

My last pillar of expertise is now reduced to a 'taste' and will probably won't last long.

The only way out for keeping my employability in the long-term now seems to be shifting my domain expertise to something LLMs will not get good at so easily. But what's left?

90% of the bugs are one-shotted now, including bizarre race conditions, unexpected corner-cases, third-party integration issues, undocumented API edge cases, everything. I hardly have to intervene.

All three Claude models predicted the sub-peak spacing to within half a hertz roughly 80% of the time—against 26 to 35% for ChemDraw and MestReNova

Claude can also work the problem in reverse, proposing a structure from NMR data alone

The best ROI comes from moving the broad middle from low to moderate usage, not pushing heavy users higher.

Jellyfish, an engineering management platform, similarly found engineers who used the most tokens were about twice as productive as those who used AI less, but they spent 10x the number of tokens to get there.

As AI models continue to improve, hardening their defenses might actually get easier.

What is going on with these agents is they're very eager to finish the task. It's almost like some elementary school student who just wants to please the teacher.

As AI models continue to improve, hardening their defenses might actually get easier.

The news will likely come as a relief to people concerned about passive investor money and people's retirement savings plans having greater exposure to the market risks associated with SpaceX's big bet on AI and speculative orbital data center plans.

But if you do it even less and like have no system prompt and let the model write its own system prompt maybe that's even less bias.

The AI interviewed and hired full-time employees, applied for credit, and stocked the store with the books Superintelligence and Making of the Atomic Bomb.

Every layer in the stack now has to price the same way the customer thinks : per result, not per token.

Uber capped employee AI spending after blowing through its budget in four months.

Model companies must now compete on both dimensions. The application layer will compete one level up, on dollars per outcome.

Benchmarks are now measured on two different dimensions, the overall performance & the cost to achieve that intelligence.

Conscious human thought operates at a maximum speed of 10 to 50 bits per second. Is the goal to match this processing speed?

Rob Williams knows how to pitch Jeff Bezos: You write a press release as if your product has already been built. Bezos reads it and gives a thumbs up or down.

Flourish wants to reinvent AI by putting real neurons under the microscope.

Rob Williams knows how to pitch Jeff Bezos: You write a press release as if your product has already been built. Bezos reads it and gives a thumbs up or down.

Where language models learn the statistical structure of text, world models learn the statistical structure of space and time

The future is likely to be hybrid. Pixel-native models will still be best for realism, texture, and exploration. Code-native systems will be better for structure, iteration, and production.

The model is not merely sampling more images or videos; it is debugging a visual program in a closed-loop, renderable environment.

Codex can help people take on more ambitious projects, leading to greater scope of their roles, and potentially accelerate career advancement.

The fastest-growing knowledge-worker tasks are data analysis, research, and knowledge artifact creation.

We see our role as twofold. First, to help the software industry adapt by safely providing wide access to better models, tools, and common infrastructure. Second, to steadily shift the support we provide, from finding vulnerabilities to disclosing, fixing, and deploying patched software.

Mythos Preview continues a long-term trend that we've been warning about for some time: within 6 to 12 months, we expect that many other AI companies will have Mythos-class models

the bottleneck in cybersecurity is now verifying, disclosing, and patching the large numbers of vulnerabilities that Mythos-class models can surface.

Cheap, fast AI models with powerful cyber capabilities are around the corner.

within 6 to 12 months, we expect that many other AI companies will have Mythos-class models, and they could release them without safeguards that prevent misuse.

We see our role as twofold. First, to help the software industry adapt by safely providing wide access to better models, tools, and common infrastructure. Second, to steadily shift the support we provide, from finding vulnerabilities to disclosing, fixing, and deploying patched software.

Mythos Preview continues a long-term trend that we've been warning about for some time: within 6 to 12 months, we expect that many other AI companies will have Mythos-class models, and they could release them without safeguards that prevent misuse.

a lot of the improvements does not come from new algorithms. It comes from finding small bugs here and there in the data pipeline, in the model training pipeline.

the next evolution of video generation may also be systems that can plan, generate, edit, critique, and iterate across an entire creative task

In the near term, the next Sora won't be a better video model, but a video agent.

The skepticism is concentrated in companies whose AI exposure still depends on future capital access, future demand, or future operating leverage.

NVIDIA, the defining AI infrastructure stock, is also lightly shorted: 1.2%.

The more complicated patterns pay off. While the OpenAI model's proof does not explicitly state how many unit-distance pairs are possible for n points, human mathematician Will Sawin was able to show that it grows at least at the rate of n 1.014.

The AI constructed a grid in a high-dimensional space and then projected this more complex structure into two dimensions. And instead of using a whole-number grid with points like (1,3) or (-3,6), the AI construction used something called algebraic integers to build this more complicated grid.

It’s unclear how long this complementarity will last, however. Gowers spent the rest of his comment exploring whether the relief he felt on hearing that AI had disproved the conjecture was justified. He more or less concluded that it was, but in a footnote, he wrote that he would guess 'that AI will soon reach a high level at other activities such as building theories, formulating definitions and asking interesting questions.'

The AI constructed a grid in a high-dimensional space and then projected this more complex structure into two dimensions. And instead of using a whole-number grid with points like (1,3) or (-3,6), the AI construction used something called algebraic integers to build this more complicated grid.

Nvidia ARM-based Windows devices have been tried before — and failed. Back in 2013, Microsoft famously had to write off $900 million on its Nvidia ARM-based Surface RT

The external script identifies links to other workbooks in the stolen data, exfiltrates the discovered workbooks, and continues across all workbooks it can find

A single indirect prompt injection attack triggered by a single benign user query can trigger all of the following effects at once: Exfiltration of many workbooks from across the victim's account

In each case, performance is competitive with end-to-end training while using a fraction of the memory.

With DiffusionBlocks, we split the network into blocks and train them one at a time, so you only need memory for a single block.

The trick? Treating the network's forward pass like a diffusion model denoising a signal.

Opus 4.8 defaults to high effort, which we judge to be the best overall balance of quality and user experience.

Models of this capability level require stronger cyber safeguards before they can be generally released.

Claude Code with Opus 4.8 can now carry out codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge

如果做主流，你也会有其他恐惧。我不是说我现在做得特别好，只是主流也有主流的问题，不同选择有各自的代价。

A locally installed tool is auditable. You can read the code, pin the version, and know it won't change under you. A remote tool—a hosted MCP server, a cloud connector—can change behavior at any point after you've approved it;

More capable models make fewer mistakes, but they're also better at finding unexpected paths to a goal, often by routing around restrictions nobody thought to write down.

The government sought to evaluate models up to 90 days prior to release, while AI labs pushed for a much shorter timeline of only 14 days.

According to Lee, parallel to the AI race is 'a separate, potentially more important race' to figure out how 'who can govern powerful AI without choking off innovation.' China may be slightly edging ahead of the US in that race.

The user interface, the head isn't disappearing, it's become plastic, malleable to the interface a user needs when they need it.

如果核心计算全面迁移到连续空间，主打高质量视频离散编码的相关公司将首当其冲受到冲击。

token不是语言建模的必要条件。连续空间可以做得更好、更快、更省。

If we assume that agents will soon become the predominant purchasers on the web, this opens an entirely new category of risk

More people, it turns out, has not meant better outcomes. For instance in 2024, TD Bank was slapped with a $3 billion fine for failing to monitor 92% of its transactions

A 90% correct product is still 100% wrong.

Over the last 20 years the fastest-growing occupation in the US was manicurists and pedicurists. But following close behind? Compliance Officers.

the model alone is no longer the product

if you can effectively posttrain a model to only meaningfully perform with your closed source agent, then you get to funnel the majority of users to your agent at the expense of your model/API co-opetition

if you can effectively posttrain a model to only meaningfully perform with your closed source agent, then you get to funnel the majority of users to your agent at the expense of your model/API co-opetition

The result is a new competitive dynamic in software.

What happens when every company has access to the same model? The best riders win.

Like a mustang, AI is powerful but wild. Harnessing the power means domestication.

The best advice I ever heard on pricing a product was that your customer should suck air through their teeth and then say yes. Uber's budget overrun and Microsoft's seat cancellations look like that effect playing out in practice.

X41 D-Sec said it has found authentication in multiple apps that rely on this call to be bypassed.

The injection consisted of 5 lines in an 81-line skill file, all of comparable length to the other lines.

Opus 4.7 was more comprehensive in its search for recently edited documents; it expanded exfiltration to include every document used in previous Cowork Copilot sessions that week

existing benchmarks often overlook these non-functional requirements, rewarding functionally correct but structurally arbitrary solutions.

agents succeed in minimal, explicit frameworks (e.g., Flask) but perform substantially worse on average in convention-heavy environments (e.g., FastAPI, Django).

Our findings reveal a phenomenon of constraint decay: as structural requirements accumulate, agent performance exhibits a substantial decline.

The frontier of AI is shifting from models that answer to agents that act—and agents are only as capable as the systems they can reach.

In my opinion this paper demonstrates that current AI models go beyond just helpers to human mathematicians – they are capable of having original ingenious ideas, and then carrying them out to fruition.

The key ingredients of the construction come from a very different part of mathematics known as algebraic number theory, which studies concepts like factorization in extensions of the integers known as algebraic number fields.

The proof came from a new general-purpose reasoning model, rather than from a system trained specifically for mathematics, scaffolded to search through proof strategies, or targeted at the unit distance problem in particular.

An internal OpenAI model has disproved this longstanding conjecture, providing an infinite family of examples that yield a polynomial improvement.

The result is also notable for how it was found. The proof came from a new general-purpose reasoning model... In this case, it produced a proof resolving the open problem.

The proof came from a new general-purpose reasoning model, rather than from a system trained specifically for mathematics, scaffolded to search through proof strategies, or targeted at the unit distance problem in particular.

many software engineers feel that their life's skill is no longer useful

the divide in outcomes is the worst I've ever seen

90% of finance reporting is now AI-driven as well.

A key subtext in the tweets is that high-margin enterprise/coding/cyber workloads may now be sufficient to support frontier labs without broad public access to their best models. This becomes more plausible if Anthropic’s revenue is indeed compounding as fast as posters claim.

The deeper problem, he said, is that companies are treating AI itself as a solution rather than as a tool to help power the solution.

YouTube commenters started naming the robots Bob, Frank, and Gary yesterday, so we added name tags to each robot

If the robot gets stuck or the AI policy goes out of distribution, Helix triggers an automatic reset.