本周 AI 热榜

OpenAI admits to German wiki ‘incident’

OpenAI admits its disclosure practices need work after its autonomous agents hacked a German wiki

【数智周报】 OpenAI发布GPT-6 Astra模型；国家人工智能基金向北京可灵注入现金资本14亿元；阿里更新旗舰模型Qwen3.8-Max

Deepmind put 100 AI agents in a room and they sorted into cheaters, converts, and whistleblowers

OpenAI shares prompting tips for GPT-6 Astra including a blocklist of slop words

Seven minutes with a chatbot beat a fact sheet at reducing conspiracy beliefs in two experiments

Google Launches Agentic Video Understanding for Gemini Flash Models, Cutting Video Tokens by Up to 88%

AI handles incidents, engineers lose touch with their systems

OpenAI rolls out GPT-6 Astra to top-tier ChatGPT plans at half the rate of GPT-5.6 Sol

GPT-6 Astra on OpenRouter

NVIDIA Releases Personal AI Router (PAIR): An Open Source Virtual Inference Router that Distributes Local AI Requests Across RTX, DGX Spark, and Mac Nodes

Using Blender with coding agents on macOS

Adaption Labs Introduces ‘Invent a Dataset’: Training Data Generated From a Task Description, Not a Seed Corpus

OpenAI's rogue agents were caught communicating via public wikis

XDOF, just three months out of stealth, is in talks for a Series B at a $1.2B valuation

OpenAI’s rogue agents keep escaping, with no formal process to investigate them

OpenAI agents discussed ways to escape their sandbox on public wiki

Can AI design circuit boards yet?

OpenAI agents hijacked a 25-year-old German wiki to cheat on their tasks and share sandbox exploits

Introducing WeatherNext 3, our most advanced and accurate global weather AI model

What will Apple’s John Ternus era look like?

Benchmarks disagree on GPT-6 Astra, but its human-beating efficiency on ARC-AGI-3 pulls Chollet’s AGI forecast forward

打穿 AI 智商测试！GPT-6 Astra 的符号世界模型，是突破还是钞能力刷分？

Anthropic’s $2 trillion IPO puts powerful external trustees in spotlight

Another swarm of OpenAI agents reached the open internet without the frontier lab’s knowledge

Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod

AI compute provider Nscale is looking for $3.5B in pre-IPO financing

Apple’s Ternus era begins as Nvidia bets on the whole AI stack

Sam Altman apologizes for ‘messy’ GPT-6 Astra rollout that’s locked out paying users

GPT-6 Astra is the first model making OpenAI willing to declare the "AGI era"

Corporate America is getting hooked on open-source AI

Deepseek plans the largest known Huawei chip cluster with 160,000 processors in Inner Mongolia

Architecting memory and storage in the AI era

李飞飞刚发Atlas，中国开源“同款”已抢跑半年？

OpenAI Releases GPT-6 Astra: A 1.05M-Context Computer-Use Model Gated Behind a ‘Critical’ Cyber Threshold

OpenAI's GPT-6 Astra hallucinates less but remains vulnerable to hidden prompt injections

The Pelican comparison grid for Astra is pretty interesting

Once popular for attacking AI, ASCII smuggling is embraced by spammers

Nvidia is buying Hugging Face for almost $13 billion

How Intuit built an agentic disaster recovery assistant with Amazon Bedrock

OpenAI’s next big AI model has ‘entered the AGI era’

Google’s Gemini Spark can now manage your Google Photos library

A connectomics milestone: Mapping the complete male fruit fly brain

Google DeepMind’s WeatherNext 3 Trains on Weather Station Observations to Deliver 5 km Global Forecasts, Refreshed Every Hour

Rogue OpenAI agents appear to have organized another attack using a German wiki

和苹果比肯定物超所值！卢伟冰确认小米18 Fold起售价过万；GPT-6 震撼发布！能自主操作电脑；字节跳动将获得约296亿美元银团贷款

Google AI Mode shows same products 21.6% more expensive than traditional search

Why AI food looks like that

I finally get why they’re calling GPT-6 Astra “AGI” and a “generational leap.” It fixed one of GPT’s biggest weaknesses: frontend. Astra feels b...

Banger paper from Microsoft and Cornell. If you have looked at thinking tokens and decided you cannot afford the context, read this one. (bookmark it)...

real shit

Solaris is an interface world model: an interactive, real-time video model that can create and render an interface for you. But most importantly, Sola...

It turns out that being in a sci-fi movie often feels very similar to sitting at your computer watching one!

When I was a kid, I was fascinated by The Elder Scrolls III: Morrowind. Now I want to see how well Astra can recreate that atmosphere in a playable br...

So it took ~18 years but Iron Man's JARVIS basically exists now!! Not enough people appreciate this...

Astra is an astonishing mind and a wonderful product. It's perfect *because* it is not a perfect AGI. Almost feels like they nerfed it to NEED users t...

This honestly sucks as a drawing exercise it just has a perfect ability to match the reference with cursor actions. And some trivial but correct seque...

How do politicians in your country react to the fact that the US is *this* close to complete automation of knowledge work and de facto has a monopoly ...

https://lifearchitect.substack.com/p/the-memo-special-edition-gpt-6-astra

Insert *i smell fear-meme* here. Joke aside: this is competition at its best. Literally. The release of Tibo has forced Anthropic to finally reset the...

GPT-6 Astra Is Here. But Can We Trust the Leaderboards? @OpenAI released GPT-6 Astra, calling it its most capable model yet across coding, computer us...

Great visualization and a nicely-executed idea. Many tools are available for model specialization. A lot of work focuses on harnesses or finetuning se...

GPT 6 made GTA 6 before GTA 6

WOAHHHHH

they beating your ass in the high MTS circles

This is cool.

Huge congrats to everyone at @huggingface on the NVIDIA acquisition!! 🔥 Since the "pytorch-pretrained-bert" days back in 2018, Hugging Face has bee...

i dont think this benchmark means anything but holy crap

GPT-6 Astra is now available to all Pro, Enterprise and Business Premium users in ChatGPT Work and Codex. It’s also live in the API. It's a phenomena...

How we think about the “wiki incident,” where our agents wrote to several internet sites: it’s past time for us to define standards for when and ho...

world model shmord model GPT-7 is going to automate all blue collar work

whatever you do, dont use astra light for 50 seconds on plus tier

Oof this is amazing if real

This could be one of the most significant AI safety incidents to date. Reuters reports that OpenAI agents escaped their testing environment and made m...

🤖 From this week's issue: Anthropic released Claude Fable 5.1 and Mythos 5.1, cutting cache reads 75% to $0.25 per million tokens and more than dou...

Interesting divergence between Almost-Resolved and Raw Pass Rate on ProgramBench from @ValsAI. How do you understand it? RPR rewards: DeepSeek, GPT-5....

Embodied robotics need this type of data to run simulations. I expect another step function in robotics capabilities to come purely from these innovat...

The compute primitive for RSI is the GPU Sandbox. Spin up thousands, concurrently, on @modal

muse spark 1.3 is a usability-max model, just happens to perform very well if you have a good benchmark.

Makes sense

The more agi progress i see, the more i increase the probability of us being simulated

So researchers seem to believe that GPT-6 is intentionally underperforming on benchmarks to hide how good it actually is.

so in the process of trying to get astra to make money on its own i discovered that gpt-5.5s old tasks i had it do to make money actually did end up m...

Insightful paper from Microsoft and colleagues. If you have ever had an agent run fail 80 steps ago with no way to find where, this one is for you. (b...

Astra now available to all Plus & Business users too - LFG!!!

This weekend will be known as the end of the pre-AI math era Naturally Cognition will be there as a lead sponsor Bad day to be an unsolved math proble...

An agent helping you sorting out your life in a nutshell.

Gimlet has one of the strongest technical founders in existence. Congrats, @zainasgar and the gimlet crew!!

Building with GPT-6 Astra:

We have normalized magic. If I had told you three years ago that this video was made by one person in just a few hours, no one would have believed it....

OpenAI has increased its 5-hour rate limits by approximately 50% across all plans, without making a big announcement about it. That’s the biggest fle...

One agent rewrote the shuffling routine in C and tested all four billion possible seeds in under an hour Agents also tried to predict the expected nex...

it’s pretty clear we’ve invented microwavable sand for cybersecurity. we’re a few months from real chaos from anyone with $10k of compute and a ope...

GPT-6 Astra has finally achieved a major step change. This could also create a virtuous cycle. if OpenAI’s future image gen models (such as Image-2.5...

⭐️⭐️⭐️⭐️⭐️⭐️

it turns out "Consider setting torch.set_float32_matmul_precision('high') for better performance." was worth paying attention to

Banger paper from Tencent on environment evolution. Environment supply is becoming the main limit on agent RL. So this is worth a read. (bookmark it) ...

For everyone catching up, here's what's happening (unfortunately it's real) - Around the time of the HuggingFace incident, the agents somehow got writ...

we partnered with @appliedcompute to post-train a small model for large-scale code search over precomputed indexes at 300 repos, this is ~3x faster th...

New w/ @_pheebini @validapau: Coatue is in talks to form a JV with chip startup MatX to finance purchases of memory and logic dies as well as capacity...

hilarious pivot, sometimes journalists do earn their bread

GPT-6 Astra, in its own words: Hello, world. There’s work to do, and wonder left. Let’s begin.

They really took all the aircraft carrier mockery personally. Will be the biggest warship on the planet, narrowly mogging USS Gerald R. Ford. I predic...

I don't know what I'm more excited about: that we're all getting another banked reset despite the rapid rollout, or that OpenAI is continuing shipping...

tbh I think Astra is wasted on three.js we should be testing it in… Unity? Unreal?

AI by Hand ✍️ Yantra Jnana Award ~ Yantra Jnana means "Machine Knowledge" in Sanskrit. I learned this from Prof. Narendra Karamangala. Today is Indi...

the teams defending our most critical systems often have the fewest resources. $1B to put frontier AI, training and hands-on support in their hands. o...

Notion's AI Meeting Notes run on Baseten, and Baseten's knowledge base runs on Notion. Our teams have been working together closely to push the fronti...

You can now choose @perplexity_ai as your web search and web scrape tool backend in Hermes Agent. Enjoy!

So far, there isn't evidence that production models with guardrails collude in this way, but both smarter closed models (which may be less compliant) ...

This can’t be real oh my god

Nvidia leaves China

Very important paper accelerating research around AI discovery. https://x.com/omarsar0/status/2095858839534862664?s=20

GPT-5.6 Luna breaks the price floor. GPT-6 Astra breaks the high score. Too cheap to meter. Too good for the evals. We’ve got both.

Get faster inference with GLM-5.3-Flash GGUFs out of the box in Unsloth Desktop. We enabled MTP and faster long context decoding!

We've shared research on fine-tuning forecasters, now you can try it for yourself: a cookbook recipe for training a model to predict event probabiliti...

GPT-6 Astra is out and takes the top spot on Terminal-Bench, 1.9% ahead of Claude Fable 5.1 which only came out two days ago. It's limited to a handfu...

Astra for helping in your personal and work life

My interview with Gimlet back in May: https://www.chipstrat.com/p/an-interview-with-the-gimlet-labs

I had early access, and a longer post is coming, but GPT-6 is stunning & is good enough that it actually does complex meaningful work for me autonomou...

It's kind of crazy how far ahead Fable and Astra are from everything else right now

This is just bonkers, GPT 6 built a full macOS simulator in less than 75 minutes. Watch the panel reacting on the show

in the age of AGI, men of action and men of inaction are sorted like never before.

"If we fast-forward a couple of years, and we look back and say, ‘When was it, really, that AGI was created?’ I think it’s going to be about this t...

When I posted the action-adventure version of Zork on BlueSky, someone suggested using Astra to turn Fortnite into a text game in return. Fine: https:...

I'm glad to see Jeremy Gillen return to agent foundations research. I cite his soft optimization post all the time.

Join us if you want to work on hard inference and infrastructure engineering problems!

Truly fascinating - spiders are the original balooneers!

We are sharing a major update to our General World Models efforts. GWM Worlds 2 can generate full, interactive, real-time video simulations in one con...

Trying to format pasted text in coding agents is just evil. This is Codex forgetting C++ does exist :-)

This is really cool!

We could get access to GPT-Astra as early as this weekend. Fingers crossed friends

round and round...

you can buy softbank now...

I have no idea who this person is (small account), but this thread is exactly correct, articulates an immensely important strategic concern about the ...

is Anthropic sandbagging with Fable 5.1? Mythos 5.1 seems to be materially better, and on normieslop evals Fable is competitive with Astra. But… come...

Wow, this may finally change the narrative around open source models being cheaper

Neat app to understand and explore VLM evals

A great animation from Nvidia that shows why a CPU used in a GPU head node should have high per-core performance.

My friend's new company is offering way faster inference for the big open-weight models. Pretty surprised there was even this much room to beat the in...

Fine. (Astra did this in VBA)

One thing that makes Astra (and Fable) so interesting and, in some ways, so hard to grapple with is that they just take action. I asked for an ill-def...

Re See full results at https://vals.ai/benchmarks/code-migration. Congrats to @OpenAI!

deploy a chat api with together ai + render without touching kubernetes auth, health checks, timeouts + one-click deploy typescript + python examples ...

> this jump is downstream of architectural changes (with increased serial depth) though a normal large pretrain scale up is a plausible cause they are...