About This Week’s Covers

This week’s covers are a tribute to Tim Curry, who died on August 25th. Dolly Parton died the same day, but she is such an amazing icon that I didn’t know how to do her justice with imagery. However, Tim Curry had so many fun characters that I decided to go with the Clue poster from 1985.

The original Clue poster is a vertical aspect ratio of a black and white spooky mansion drawn in the style of Gothic pen and ink with the textured look of cross-hatching and stippling without using grayscale. It has a little bit of a wood-style look to it, like the 19th-century mystery novels.

Edward Gorey is probably the most famous pioneer of this cross-hatched look, but there are “three” other names to know.

1. Gustave Doré: A French artist who pioneered dark, detailed wood engravings for gothic and epic literature (like Edgar Allan Poe’s The Raven and Dante’s Inferno).
2. Victorian “Penny Dreadful” Illustrators: Anonymous artists who created cheap, sensationalist horror and mystery serials in the 1800s using high-contrast woodcut and engraving techniques so they could be printed on low-quality paper.
3. Franklin Booth: An early 20th-century American illustrator who mastered the technique of making pen-and-ink drawings look exactly like traditional wood engravings through meticulous line work.

For this week’s cover themes, I used GPT Image to transform the vertical image to a horizontal one, and I swapped frontier model leaders in for the Clue characters. In the windows you see Demis Hassabis, Sam Altman, Dario Amodei, Bill Gates, Elon Musk, and Jensen Huang, and in tribute to Tim Curry, I kept the butler in all the images.

In the main cover, it’s literally Tim Curry, but in all the other covers it’s a representative image of a butler. The category cover themes are twists on the spooky mansion incorporated into the category using a Claude skill and the Gemini image generation API.

These are some of the best so far. Claude even gave them catch phrases. I’ve put a few of my favorites below:

This Week’s Humanities Selections

The humanities reading this week is a bit of playfulness from Dave Morice, also known as Dr. Alphabet, who died on August 21st this year. He used poetry in a visceral way as almost a public stunt. He would do all sorts of performance art.

One time he typed a thousand poems in twelve hours in a bookstore. Another time, in 1977, he spray-painted a poem on a paper roll stretched from goalpost to goalpost during a high school halftime ceremony. He also invented a fake poet named Joyce Holland, whose magazine Matchbook sold one-word poems for a nickel, printed on one-inch pages, stapled inside donated matchbooks.

This is an absolute antidote to artificial intelligence. It’s also a fun throwback to the age of typewriters, since a lot of Dave Morice‘s performance art was created by the physicality of the typing machine. I’ve put a few examples below.

This week’s song is one of my favorite clips of Kenny Rogers singing live with Dolly Parton on We’ve Got Tonight.

His reaction to Dolly is a classic as her voice cuts through the quiet, and he visibly shows how much he adores her… They have some kitschy 1985 banter, which is scripted but still well delivered by two consummate artists.

As luck would have it… it’s perfect because We’ve Got Tonight live in 1985 is the same year as the movie Clue… which inspired the covers. Fair warning: this one will hit you in the feels!

This Week By The Numbers

Total Organized Headlines: 576

This Week’s Overview

This week I organized 576 links. 139 of them ended up contributing to a whopping 85 top stories, which is intimidating for me to summarize them and also you, of course, to read them all.

If you don’t want to read my summaries, you can scroll to the bottom, and there is an organized pile of links that you can skim through without any commentary from me. Hopefully that creates a nice balance if you want to be surgical or if you want to go along for the ride…

I’m eight weeks behind because I took the entire summer off, including the month of May to be with my family. I appreciate everyone who is reading this in a time-shifted way as I work my way to catch up.

Did I touch any grass this week!?
There was a lot touching grass over the past few months, but now that we’re at the end of August, you’re going to see the “Touching Grass Section” gets a lot shorter….

The week of August 28th, I went for a sunset 5k run on the beach and swam in the ocean.

I also did a nighttime Pat’s Run, which is 4.2 miles in honor of Pat Tillman. I have some personal connections to Pat Tillman, and I use him as a reminder of the best and worst of humanity. Pat was an NFL player who gave up millions to serve in the Army. He was also a Renaissance man who read the Stoics, Emerson, and Homer, and he was not afraid to be intellectual. Sadly, he was killed by friendly fire, and tragically the government chose to do the wrong thing and cover up his death to preserve the “hero’s story”.

Whether it’s Gödel’s incompleteness theorem or Pat Tillman, I enjoy holding more than one thing at a time in my mind. If you focus on the close-up, the distance becomes unfocused, and if you focus on the distance, the things right in front of you lose focus. To empathize, you have to let go of your focus and them at their focal point. A quick personal Pat’s Run is a moment to soak up all of humanity rolled up in one four-mile run… by myself.

A Pat’s Run 4.2 miler at night is what the doctor ordered

As luck would have it, this week I listened to a podcast featuring a friend of Pat’s who served with one of my best friends. Chris McCoy’s story is off the charts and he’s a world class graffiti artist . Here are three episodes worth listening:

From Streets to Special Ops: Chris McCoy’s Insane Pat Tillman Friendship Revealed
https://www.youtube.com/watch?v=5l-zUaSk5vQ

The Street That Broke Him, The Judge That Saved Him | Chris McCoy Part 2
https://www.youtube.com/watch?v=0GxlXwtn7Cs

Chris McCoy: He Served With Pat Tillman, Then Left War for the Deadliest Job on Earth
https://www.youtube.com/watch?v=S5H_9VieDH0

Let’s Do The AI News of the Week!

This week’s news is organized by topic and frontier lab. I’ve put the most interesting stories at the top, and the rest are sorted by category.

This is the most meaningful week of announcements since the launch of ChatGPT.
We’ve seen a linear progression for the past three years leading towards a complete consumer shift from chatbot to agency. I think the vast majority of the public still thinks of AI as a chatbot. But this is the week when everything shifted to consumer agents with computer use, browser use, tool use, the ability to fill out forms.

Almost like when your eyes can tell that the lighting has changed in a restaurant. Or when the sun is setting, and you suddenly realize it’s night versus dusk. That’s what happened the week of August 28th, 2026. This is the week chatbots changed to mainstream agents. Almost everything you read or watch below is going to be world-changing…

But first… I want to start with some publishing headlines to set the tone before we move into the broader AI news.

Publishing
Digiday reported on the impact of AI overviews on advertiser web traffic, and it looks like traffic dropped 10%. However, conversion rates increased.

How AI search has impacted the web traffic of over 50 advertisers
https://digiday.com/marketing/in-graphic-detail-how-ai-search-has-impacted-the-web-traffic-of-over-50-advertisers/

USA Today is working to reformat its website so that AI robot audiences will be drawn to it and result with licensing deals.
https://digiday.com/media/usa-today-co-is-reformatting-content-to-attract-more-ai-licensing-deals/

Dr. Dre on AI “I think the only people that see it as a threat are the people who have trouble creating”
https://www.nytimes.com/2026/08/23/business/jimmy-iovine-dr-dre-beats-usc.html

Google’s latest spam update reportedly targets AI-generated SEO content
https://www.searchenginejournal.com/reports-indicate-googles-spam-update-focused-on-seo-ai-content/586978/

OpenAI came out with a video about how to help your website become more agent-ready using a new standard called WebMCP. No more relying on “browser use” slowly navigating.

Ethan Mollick remarked that his spam is rising because the AI bots are now fluent enough in the humanities to comment on niche online content, conversations the more provincial robots of the past could not join.
https://x.com/emollick/status/2091720399406923882

Billionaire hedge fund chief Stanley Druckenmiller was busted using AI to write a Wall Street Journal op-ed attacking the White House. It was pretty sloppy AI usage, beyond just proofing grammar and spell check.

Hedge fund boss Stanley Druckenmiller used AI to write White House attack article
https://www.telegraph.co.uk/business/2026/08/25/hedge-fund-boss-used-ai-to-write-white-house-attack-article/

The Economist launched a personalized newsletter called The Economist for You (get ready, Nathan Fielder), which creates a reading list using the publisher’s in-house tool.

Introducing our latest newsletter, The Economist for You
https://www.economist.com/international/2026/08/18/introducing-our-latest-newsletter-the-economist-for-you

The New York Times launched an AI search tool that’s basically an interactive search engine.
https://www.semafor.com/article/08/23/2026/new-york-times-tests-out-ai-generated-search-summaries

Google Commits to Fluid Content
Rowan Cheung, one of the biggest AI newsletter publishers, interviewed Google’s Sundar Pichai, and Sundar said that in three years any content can go from any modality to any modality. It doesn’t matter how the content was originally created. People will consume it in many forms.
https://x.com/rowancheung/status/2092276652944593257

This lines up to a much earlier statement from Google that they wanted to transform the way people study and the definition of textbooks by making textbooks more fluid. I’ve been laser focused on this concept, and it’s starting to come to a head recently.

Google Statement from October 3, 2025 – a slide from my AP Presentation in June

“Textbooks are a cornerstone of education, but they have a fundamental limitation: they are a one-size fits-all medium. Any new material or alternative representation requires arduous human effort, so that textbooks cannot be adapted in a scalable manner. We present an approach for transforming and augmenting textbooks using generative AI, adding layers of multiple representations and personalization while maintaining content integrity and quality.”

Personalized, dynamic, on-demand instruction from a ‘corpus’ of knowledge (all of physics – restated at your level of interest/ability). People love to do this already:

A few weeks ago at the Local Media Association’s Annual Summit, Nikita Roy reiterated that the container is not the journalism.

There’s a great Marshall McLuhan quote about this.

“The ‘content’ of any medium is always another medium. The content of writing is speech, just as the written word is the content of print, and print is the content of the telegraph.” – Marshall McLuhan 1964

One of the best examples of McLuhan’s point… is the way that the iPhone consumed hardware and can move from modality to modality in the phone itself.

Back in June 2025, Google started to roll out data visualization features in their AI search results that automatically create interactive graphs for stock and mutual fund queries. These visualizations are built on the fly using Gemini’s reasoning capabilities to build whatever charts or graphs “Google thinks” best fits the data question.

This is bigger than it might initially seem as user interfaces will simply diffuse in front of us rather than being built in advance as fixed experiences and navigation paths. Dynamic content → dynamic interfaces.

More Publishing News
Google announced that they’re removing AI-generated search engine optimization content from search rankings.

Reports suggest Google’s latest spam update was aimed at removing AI-generated SEO content from search rankings
https://x.com/pressgazette/status/2092632887434625040

Anthropic adds built-in browser to Claude for autonomous web agents
Claude Cowork gets a built-in browser: nothing to install | Claude by Anthropic
https://claude.com/resources/articles/cowork-built-in-browser

Hermes Agent can browse the web using your Chrome profile
Browser Automation | Hermes Agent
https://hermes-agent.nousresearch.com/docs/user-guide/features/browser#real-profile-browsing-use-your-own-logins

Bill Gates Op-Ed
The choices we make about AI now are critical | Bill Gates
https://www.gatesnotes.com/work/make-ai-work-for-everyone/reader/a-turbulent-ai-era-and-critical-choices-to-make

AI will either be the greatest equalizer ever invented, or the worst source of injustice.

Even under the best circumstances, the transition to this new AI era will be one of the most turbulent times in human history. How will we use this technology to make the world a fairer place and keep it from widening the divide between rich and poor? How will we protect the people who are most vulnerable to the harms caused by artificial intelligence, including those who lose their livelihoods and the sense that they are in control of their future?

Unfortunately, right now we are not preparing for it. I don’t see evidence that leaders, experts, and communities are confronting the challenges adequately. There is no plan to ease the entry into the AI era.

I’m especially worried about young people, who will enter a workforce with fewer entry-level openings. They understand the challenge because they are the most active users of AI and see both the capabilities and the rate of improvement. It’s no wonder that so many of them feel negatively about AI.

Bill Gates is deeply worried about AI, and he’s no longer staying quiet | The Verge
https://www.theverge.com/ai-artificial-intelligence/984923/bill-gates-is-deeply-worried-about-ai-and-hes-no-longer-staying-quiet

OpenAI news
This is the week of the consumer agent explosion.

Brace yourself for TONS of examples of agents doing everyday time-saving tasks this week. Regular soul-sucking chores using not only web browsing, but now also using built-in browsers that show the user the work being done, but that work in the backend inside the agent’s virtual environment.

Anthropic, OpenAI, Nous Research, Perplexity. Almost everyone is launching web browsing and computer use agents this week.

OpenAI released a dozen YouTube tutorials, and I’ve handpicked a few of my favorites.

ChatGPT Work can now use its computer and browser to sign in to websites on web and mobile, without ChatGPT ever seeing your username or password. That means you can ask it to: • Set up utilities for a new apartment • Book a DMV or passport appointment • Check reimbursement”
https://x.com/ChatGPT/status/2092366554965107164

ChatGPT Work did, in fact, book me a haircut for 90 minutes ago and it did work I did get a haircut”
https://x.com/mweinbach/status/2093037163872915923

OpenAI’s videos can lean toward stereotypical Silicon Valley cringe, but I am giving them a pass on this one because we have to understand the concepts. Even if examples are not perfect, we need to have creativity to find the utility for ourselves…. there are a LOT this week…

Hey ChatGPT, Let’s create a website to split expenses on a tech bro ski trip. Let’s ensure we ruin the atmosphere on the mountain… Lacrosse Finance Segue Pipeline Goldfish Memory!

Legal News of the Week
This week marked three legal announcements worth knowing.

Gemini For Legal
The first is Google introducing Gemini Enterprise for legal. This includes skills designed specifically for legal work, like contract review and redlining, regulatory scanning, legal research.

Introducing Gemini Enterprise for Legal | Google Cloud Blog
https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise-for-legal

They bill it as: this is where a firm’s institutional knowledge becomes something the platform can execute rather than something a partner has to explain. AKA, see you later, paralegals. It will be interesting to see how mentorship and apprenticeship fits into understanding domains like legal if all the grunt work is now being picked up by the robots.

The Gemini system also includes connections using MCP to all sorts of document management systems, case repositories, and research services. And the third piece you might be able to predict is agents that can act within all of these data environments and handle the work: drafting, research, and screening.

Harvey Tenet
The second announcement this week is from Harvey. Harvey introduced a long-horizon legal agent using its own in-house model that’s essentially a post-trained version of Kimi K3.

Harvey Introduces Harvey Tenet: A Kimi K3 Base Post-Trained with Fireworks for Long-Horizon Legal Agent Work – MarkTechPost
https://www.marktechpost.com/2026/08/23/harvey-tenet-post-trained-kimi-k3-legal-agent-model/

“Harvey is a great example of how American companies are building world-class specialized models: they took an open-source base (Kimi K3), post-trained it on legal data, and delivered state-of-the-art performance on legal benchmarks at a fraction of the cost of frontier models.”
https://x.com/DavidSacks/status/2090790063047168473

This is another blow to the American frontier labs, as Kimi is one of the top open-source Chinese models. Harvey knows that the bottom line is what’s important for American corporate P&Ls, so they put their money where their mouth is and built a world-class model using open source, which can now give “state-of-the-art performance on legal benchmarks at a fraction of the cost of the frontier models.”

Thomson Reuters
Third this week, Thomson Reuters put out a blog explaining how they built their proprietary system called Thomson. In late July, Thomson Reuters put out benchmarks of its model, ranking as some of the top in the world. And I guess this week, Tenet is Harvey’s strong reply to this Thomson Reuters announcement. Thomson Reuters also leveraged open-source AI. Spoiler alert: they used Qwen.

How we built Thomson – Thomson Reuters Innovation Blog
https://www.thomsonreuters.com/en-us/posts/innovation/how-we-built-thomson/

Big news for Hugging Face
Hugging Face’s annual revenue hit $150 million. But even bigger news, Nvidia is in talks to purchase Hugging Face for $13 billion. Stay tuned next week.

Exclusive: Hugging Face Annualized Revenue Jumps 50% to $150 Million — The Information
https://www.theinformation.com/briefings/exclusive-hugging-face-annualized-revenue-jumps-50-150-million

Hugging Face explores a $13B sale, nearly triple its 2023 valuation
https://runtimewire.com/article/hugging-face-explores-13-billion-sale

Hugging Face Could Be Acquired for $13 Billion Amid AI Boom – Business Insider
https://www.businessinsider.com/hugging-face-could-be-acquired-13-billion-2026-8

Nvidia Agrees to Buy Open Source AI Platform Hugging Face For $12.9 Billion — The Information
https://www.theinformation.com/articles/nvidia-agrees-buy-open-source-model-repository-hugging-face-12-9-billion

Nvidia has been releasing very strong open source models for as long as I’ve been tracking headlines. IMO, they’vs overtaken every U.S. frontier lab as the best of breed creator of the full suite open source U.S. models, whether it’s robotics, OCR, transcription, vision.

The only others I can think of are Google Gemma, a few from Microsoft, Allen AI’s Molmo, Meta’s SAM family. But for consistency across every domain and skill, I’d say Nvidia is the open-source volume leader.

Nvidia wants to sell chips, and if you can run open source locally on Nvidia silicon, then you have infinitely free tokens. Hugging Face is the largest repository of open source models. A strong open-weight ecosystem helps ensure that computing demand is distributed away from the closed-source labs.

This move is a hedge against the closed-source labs like Anthropic and OpenAI, who have been investing in custom chips… TPUs, Trainium, and Jalapeno, etc… trying to reduce dependence on NVIDIA.

Hugging Face has over three million models and 18 million developers participating, and it gives NVIDIA immediate access to all of them. NVIDIA has a front row seat to open source, and can understand all of the architectural shifts, model download velocities, and all of the developer runtime trends, weeks before the general public.

The acquisition reminds me of Stripe buying OpenRouter just last week… OpenRouter decides which model should pair for which applications and task.. essentially becoming the tastemaker/bouncer for model selection and token consumption. Nvidia is building a competing marketplace using the entire Hugging Face repository and pairing it directly to their proprietary computing power.

Z.ai GLM 5.3 Crushes Costs v. Performance
The next top story may not be high on the consumer radar, but this is a story you need to be following this week.

GLM-5.3-Flash: Frontier Intelligence, Flash Cost
https://z.ai/blog/glm-5.3-flash

Z.ai released GLM 5.3 this week. There are two big things to know:

One is GLM 5.3 is extremely competent, and it’s a flash model, which is a smaller, faster version of the best models. But Z.ai’s flash model is running the same tasks as Fable 5… and it’s five times cheaper.

Two is GLM is running completely on Chinese chips. This is a shot across the bow for American chip manufacturing: Nvidia, Google, Amazon.

China is running open source on their own hardware, and they’re coming in at incredibly cheap prices.

“Give GLM-5.3 and Fable 5 the same $100 budget, and GLM-5.3 gets over 5x as much work done. On DeepSWE, that works out to about 17 solved tasks with GLM-5.3 vs. 3 with Fable 5, even though they perform almost the same on the first try.”
https://x.com/togethercompute/status/2091711899704385740

Anthropic Numbers
Anthropic is telling investors that they see $30 trillion in potential revenue.

Exclusive | Anthropic Expected to Tell Investors It Sees Over $30 Trillion in Potential Revenue – WSJ
https://www.wsj.com/tech/ai/anthropic-expected-to-tell-investors-it-sees-over-30-trillion-in-potential-revenue-a611efea

$45 Billion Anthropic Nscale Deal
Anthropic signed a $45 billion cloud deal with a company called Nscale, yet another company that I’ve never heard of, despite trying my best to follow the news for three years.

Anthropic and Nscale strike $45 billion cloud deal, sources say
https://www.cnbc.com/2026/08/26/anthropic-and-nscale-strike-45-billion-cloud-deal-sources-say.html

Nscale’s background makes for a great game of buzzword bingo, but they’re basically data centers.

Nscale is similar to CoreWeave and part of a sector called the Neo Cloud. The Neo Cloud is just a funny word for ” data centers optimized for AI training”.

Their founder, Josh Payne quit high school in Australia to work in the coal mines. During breaks, he read business books, taught himself web development, and started selling fitness gear and protein supplements (of course).

He ended up building a recruitment marketplace for construction labor , and somehow he founds an energy company using low-cost hydroelectric power for Bitcoin mining. He creates to Nscale as a source for cryptomining computing power, and now he’s pivoted the chips to AI infrastructure.

The Former Coal Miner in the Middle of the A.I. Data Center Boom – The New York Times
https://www.nytimes.com/2026/03/09/business/nscale-ai-data-center-boom.html

Anthropic poaches Google’s TPU founder to build custom AI chips
Anthropic hires Google TPU veteran Amir Salek for its own chip push
https://runtimewire.com/article/anthropic-hires-google-tpu-veteran-amir-salek-chip-team

Anthropic lets outside researchers mine real Claude conversations privately
Enabling independent research on how people use Claude \ Anthropic
https://www.anthropic.com/research/enabling-independent-research

Anthropic unifies Claude chat and Cowork memory by default
Claude’s memory works everywhere, and you decide what’s in it | Claude by Anthropic
https://claude.com/resources/articles/claudes-memory-works-everywhere-and-you-decide-whats-in-it

Anthropic merges Claude chat and Cowork memory, on by default
https://thenextweb.com/news/anthropic-claude-cowork-shared-memory-default

Anthropic launches Model Hardware Standard to benchmark AI-controlled lab equipment (robot arms, microscopes, etc)
Previewing the Model Hardware Standard \ Anthropic
https://www.anthropic.com/news/model-hardware-standard-research-preview

Salesforce embeds inside Anthropic’s Claude as SaaS replacement fears mount: Claudeforce is “a dynamic interface that thinks, reasons, and acts”

Salesforce, Anthropic expand partnership amid ‘SaaSpocalypse’ concerns
https://www.cnbc.com/2026/08/26/salesforce-anthropic-partnership-claudeforce.html

OpenAI announces its own chip: Jalapeno
“Since announcing Jalapeño, our first custom inference chip, we’ve been testing it and the system around it. The results show a major advance: more intelligence from every watt and faster responses, delivering both higher throughput and lower latency in one architecture without…”
https://x.com/OpenAI/status/2092300846675505602

Jalapeño’s first results show industry-leading speed and efficiency in AI inference | OpenAI
https://openai.com/index/jalapeno-first-results/

The full stack behind abundant intelligence | OpenAI
https://openai.com/index/the-full-stack-behind-abundant-intelligence/

“OpenAI: “Today, we shared the first measured performance results from Jalapeño, OpenAI’s first custom inference chip. On InferenceX, a public benchmark using GPT‑OSS 120B, Jalapeño delivered more peak throughput per kilowatt and lower token latency than the commercial systems in…”
https://x.com/firstadopter/status/2092266927377039838

OpenAI hires dedicated advertising lead
OpenAI is sharpening its focus on enterprise advertisers
https://digiday.com/marketing/openai-is-sharpening-its-focus-on-enterprise-advertisers/

AI advertising’s measurement problem is a governance problem
https://digiday.com/media-buying/ad-tech-briefing-ai-advertisings-measurement-problem-is-really-a-governance-problem/

OpenAI’s Codex hits 20 million users

Following Greg Brockman’s blog last week: Sam Altman warns of imminent surge in AI cyberattacks
“this is a critically important moment for cyber defense with AI; there is not much time to act. we are happy if you want to work with us or any of our competitors or partners, but please take this moment seriously. only an urgent and intense collective response will work.”
https://x.com/sama/status/2093060670472241368

A call for collective action on cyber defense | OpenAI
https://openai.com/collective-cyberdefense/

More Information Surfaces About The OpenAI -> Hugging Face agent breach

The Hugging Face incident and the road ahead | OpenAI
https://openai.com/index/hugging-face-incident-and-the-road-ahead/

Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident – METR
https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/#core-takeaways-about-this-incident?utm_source=tldrai

Waymo has 100% marketshare of California’s robotaxi rides, heads to Vegas
The Road to Autonomy® on X: “100% of California’s Paid Robotaxi Rides Belong to Waymo” / X
https://x.com/RoadToAutonomy/status/2091528222391329203

“Rolling into Vegas in style. 🎰 @VenetianVegas has named Waymo as its first official ride-hailing partner. Through this multi-year partnership, guests traveling to and from The Venetian Resort will enjoy seamless autonomous rides throughout the city.”
https://x.com/Waymo/status/2092658344720810010

Google’s $1.5 billion deal with Mechanize
This is the month of companies I’ve never heard of… here’s another.

Google is nearing a $1.5 billion deal with Mechanize. It might be an acquihire, similar to how Google paid $2.4 billion to get the CEO of Windsurf and the co-founder into DeepMind.

Mechanize is just over a year old… and they raised $9 million in April at a $500 million valuation, so a $1.5 billion deal with Google is not too shabby.

Google Reportedly in Advanced Talks for $1.5 Billion Deal With AI Coding Startup Mechanize – gHacks Tech News
https://www.ghacks.net/2026/08/26/google-reportedly-in-advanced-talks-for-1-5-billion-deal-with-ai-coding-startup-mechanize/

Mechanize was founded by Tamay Besiroglu, who is also the co-founder of Epoch AI, which is one of my favorite AI research companies and benchmark trackers. Tamay also co-created FrontierMath, and he has a master’s in economics from the University of Cambridge. The Epoch team is very strong, and it’s going to help Google quite a bit if Google can stay out of their own way.

Mechanize builds virtual work environments that let agents to kick the tires on white-collar office automation to replace human labor.

Mechanize creates all sorts of simulated workplace environments that include fake company code bases, emails, and spreadsheets, and Slack conversations… with open-ended problems that exist in real offices.

Companies who build agents can use these environments to test how well the agents will do on tough, real-life, long-horizon human tasks. These are essentially reinforcement learning gyms… digital office test tracks.

The main benchmark measures how well an agent can navigate a desktop ecosystem: working with office software suites, spreadsheets, editing documents, going through internal databases and searching CRM tools, using a browser get information, fill out forms, and doing administrative work.

These simulations test the ability of agents to work together dynamically to coordinate and communicate with each other using email and Slack to request information from their coworkers, which are either simulated bots or simulated humans. The simulation includes interruptions, reprioritizations, and intentionally introduces new obstacles and curveballs.

The “dojo” might have an agent’s computer crash, or require an agent to shift focus many times, switch to a different goal, and then come back to their original goal or project without losing any of the meaning.

It makes sense that Google would want to invest in Mechanize, since Google’s office suite covers all these areas of work. A virtual office gym puts Google in a powerful position against Microsoft in particular (it’s wild how you never see Apple in any of these discussions).

Double Blind Evaluations
Google announced a double-blind evaluation for frontier models. One of the most important things in testing a frontier model is ensuring that the test questions haven’t leaked, so that the frontier model cannot study on the test. A double-blind evaluation guarantees that the testing is secure and the model cannot train on or improve from taking the test .

Piloting the world’s first double-blind AI evaluations — Google DeepMind
https://deepmind.google/blog/piloting-the-worlds-first-double-blind-ai-evaluations/

WeatherNext: Google’s AI model for forecasting cyclones
Google AI on X: “WeatherNext: Our AI model helping to forecast cyclones” / X
https://x.com/GoogleAI/status/2092275116503707733

Google launches Gemini 3.5 Transcribe with 2.6% error rate
Introducing Gemini 3.5 Transcribe
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5-transcribe/

Nvidia news of the week

Nvidia hits industry-first $108 billion quarter
NVIDIA’s $108b Quarter • Tomasz Tunguz
https://tomtunguz.com/nvidia-q2-fy27-earnings

SpaceX and Nvidia plan orbital AI supercomputer launch in 2026
“The first CPU built for agents is going to work at scale. @SpaceX is deploying NVIDIA Vera to accelerate the orchestration, code execution, and data processing that powers its next generation of agentic AI — keeping GPUs fed and agents acting fast.”
https://x.com/nvidia/status/2091920680317046847

Elon Musk on X: “SpaceX, in partnership with Nvidia, has designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028”
https://x.com/elonmusk/status/2091939113008238838?s=20

Nvidia in talks to back Perplexity at $30 billion valuation
Nvidia Discusses Perplexity Investment at $30 Billion-Plus Valuation — The Information
https://www.theinformation.com/articles/nvidia-discusses-perplexity-investment-30-billion-plus-valuation-considered-tech-licensing-deal

Yutori launches 27B computer-use model at fraction of rival costs
Introducing Navigator n2: 27B Computer-Use Model | Yutori
https://yutori.com/blog/introducing-n2

Yutori is a US company based in San Francisco. The founder, Abhishek Das, got his PhD at Georgia Tech and did work on agents many years before they were cool. He worked at the fundamental AI research lab, FAIR, at Meta, famously run by Yann LeCun, where he helped start the Open Catalyst Project, which is now called FAIR Chemistry. The other co-founder is Devi Parikh, also from Meta at GA Tech. She was the senior director of generative AI at Meta and an associate professor at Georgia Tech. The third co-founder and chief scientist is Dhruv Batra, also from Georgia Tech and also from FAIR at Meta. So a pretty close trio of friends who have traveled together through their past two jobs.

Yutori’s launched a proprietary computer use model family called Navigator that can use multimodal understanding to see the pixels and interpret screens and execute actions.

This seems like a very competitive space, considering how much is coming out from every other lab for the past two years: from ByteDance to Alibaba to Amazon to Microsoft to Nvidia, Anthropic, OpenAI, Gemini, Qwen. I can’t even think of any model that doesn’t have computer use.

The differentiator is a low price, while still outperforming Claude Opus 5, GPT 5.6, and Claude Fable.

Personal AI agent Instinct hits $2.5 billion valuation after viral launch
Exclusive | The Latest Viral AI Assistant Rocketing Across Silicon Valley – WSJ
https://www.wsj.com/tech/ai/the-latest-viral-ai-assistant-rocketing-across-silicon-valley-abb46276?mod=author_content_page_1_pos_1

Instinct
https://instinct.com/

Instinct is also a San Francisco-based AI company, founded in late 2025 by a 23-year-old named Noah Shinn. Noah authored an influential AI research paper called Reflexion. It pioneered a loop technique that let agents evaluate their own output and self-correct. His research was such a hit that he dropped out of Northeastern and joined a company Sierra, run by former OpenAI chairman and Salesforce co-CEO Bret Taylor. Noah left Sierra in September 2025 to create his own consumer product.

Instinct stayed in stealth until April 2026 and then opened up with an invite-only beta in August, with five invites per person. People started selling invites for hundreds of dollars on eBay. By the end of August, Instinct was valued at $2.5 billion, with a $250 million seed round. Just mind-blowing, but it makes sense given his background.

I have Invites, so reach out if you want one.

NVIDIA’s AVO agent scores perfect 100% on ARC-AGI-3 benchmark
NVIDIA has an agent harness called AVO. We’ve talked a lot about harnesses. This one runs using Claude Opus 5. This is a great example the value of a harness…

On its own, Claude Opus 5 only scores a 30% on the ARC AGI 3 benchmark. But when run through NVIDIA’s AVO system, Claude Opus 5 got a perfect 100% .

Perplexity and Nvidia launch local AI agent with zero token costs
Earlier we talked about NVIDIA’s investment in an equity stake in Perplexity. Well, this week Perplexity launched Portable Computer, which runs on local DGX Spark hardware. It’s a fully local AI agent that has zero token costs. But of course, as we just learned, it has to be running on an agentic computing stack like NVIDIA’s DGX Spark. You can get a 64 gigabyte configuration on a Spark workstation for about $5,000. And if you go up to a four terabyte cluster, it costs about $14,000.

Introducing Portable Computer
https://www.perplexity.ai/hub/blog/introducing-portable-computer-for-local-first-ai

Perplexity partners with Nvidia to launch Portable Computer, a fully local AI agent with zero token costs | VentureBeat
https://venturebeat.com/infrastructure/perplexity-partners-with-nvidia-to-launch-portable-computer-a-fully-local-ai-agent-with-zero-token-costs

Silico, an AI interpretability agent, authored an original research paper
Goodfire is an interpretability company, and Silico is an automated interpretability agent created to help researchers guide massive models.

Interpretability is one of the most important and least understood (by laypeople) disciplines in artificial intelligence. Interpretability is the study of going back and looking inside AI after it’s trained to understand how the neural networks work and adjust things, almost the way maybe a neurosurgeon would. It’s the study of the proverbial black box.

In this case, Silico achieved a major milestone by acting as an autonomous researcher. Eric Bigelow is a member of the technical staff at Goodfire, and he was working on a paper called Forking Paths. It was a follow-up to a 2024 paper Eric wrote about how large language models make decisions and how to measure uncertainty by tracking the forks that occur as they generate their reasoning for any given answer.

Silico did all the research for the follow up paper in under a week. Silico ran all the experiments, gathered the data, and handled the execution, enough so that it could be listed as the first author of a paper. That’s the typically junior researcher or a PhD student who does all the heavy lifting and runs the experiments on behalf of the researcher. So Eric Bigelow was able to use Silico as his wingman, and he compared it to having a highly skilled first- or second-year PhD student. https://arxiv.org/abs/2608.19611

Goodfire on X: “The research in this paper was performed by Silico, our interpretability agent. @EricBigelow: “it was roughly on a similar level as working with a first- or second-year PhD student. It’s wild to think that if Silico were a person, they’d be first author and I’d be last author on” / X
https://x.com/GoodfireAI/status/2092661172680245602

Keenable raises $26M to build web search index for AI agents
Keenable briefly made an appearance in a tweet last week, where the Artificial Analysis Search Index benchmark was announced, and Keenable was part of a list of companies that flew under my radar.

“Announcing the Artificial Analysis Search Index, benchmarking how search API providers perform on quality, cost, and speed when used by an agent. We are initiating coverage with Parallel, Exa, Firecrawl, You.com, Tavily, Keenable, and Brave Search.”
https://x.com/ArtificialAnlys/status/2089755262915936661)

Accel-backed Keenable is indexing the web for AI agents | TechCrunch
https://techcrunch.com/2026/08/25/accel-backed-keenable-is-indexing-the-web-for-ai-agents/

Alibaba launches Wan3.0 AI video model after record $10 billion share sale
The new software can generate 30-second videos directly from text documents, data spreadsheets, presentation slides, and web pages.
https://finance.yahoo.com/technology/ai/articles/alibaba-launches-wan3-0-ai-131300534.html

Deepseek Vision Model
Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks
https://the-decoder.com/deepseek-releases-experimental-flash-vision-model-that-rivals-opus-4-8-on-agent-benchmarks/

World-first AI-assisted brain surgery preserves patient’s vision
First patient in live AI assisted sight-saving brain surgery
https://www.uclh.nhs.uk/news/first-patient-live-ai-assisted-sight-saving-brain-surgery

Mistral strikes hundreds-of-millions euro sovereign AI deal with Saudi HUMAIN
Mistral is the preeminent French open source lab, founded by two former Meta employees, and the current CEO was a research scientist at Google DeepMind. One of Mistral’s core missions is to help build sovereign AI systems for governments, banks, and corporations… where the sponsor state or org has complete control over the AI, the data, the alignment, the operations.

This week, Mistral partnered with HUMAIN, which was founded by Mohammed bin Salman, the Crown Prince of Saudi Arabia, using the country’s massive public investment fund. The CEO of HUMAIN is Tareq Amin, who spent most of his career at Rakuten as well as Huawei and T-Mobile.

Mistral x HUMAIN | Mistral
https://mistral.ai/news/mistral-x-humain/

HUMAIN is a state-backed tech company tasked with turning Saudi Arabia into a global superpower in AI. They’re trying to build not just an app or a single model, but a sovereign technological ecosystem for the benefit of Saudi Arabia.

This could be physical infrastructure like data centers, or software models to compete with U.S. frontier labs. In particular, their niche is Arabic-first AI models trained on billions of Arabic words and concepts to natively understand Arabic, as opposed to most of the frontier models, which are trained first on English.

The partnership will have Mistral working with HUMAIN to build sovereign AI in Saudi Arabia for the broader Middle East. HUMAIN will build the giant hyper-localized data centers in Saudi Arabia, and then Mistral deploy highly advanced AI models that can be dropped directly into these local data centers so that the data never leaves the Middle East.

Meta launches Muse Image, an agentic text-to-image model priced at $0.01 (seems like diffusion + Segment Anything)
This week Meta launched Muse Image, which they call an agentic text-to-image model, priced at a very cheap one cent per image.

Muse Image | Meta
https://dev.meta.ai/models/muse-image

At first glance, it seems like Meta is using their segmentation technology to power the image tool. Segmentation is when you can identify objects or the boundaries of objects within images or videos.

In Photoshop, this would be called masking. For image editing, that would be very powerful because it can isolate the locations of an image that you want to change, so you don’t have to re-render and possibly lose the render that you got the first time.

I’ve always been under the impression that ChatGPT and Gemini both use segmentation to some extent when they edit images, so I’m not sure this is a completely new thing. But the price that Meta is charging is very cheap, and I wonder if it’s because their open source model Segment Anything is able to power the backend at a very inexpensive rate, or perhaps they’re just eating the cost.

Ex-Prometheus AI founders launch startup to model the universe
Accelerated Understanding was founded by Anima Anandkumar, a Caltech professor and a former NVIDIA executive, and Benedikt Jenik, an MIT grad and research assistant.

The duo is famous for declining multimillion-dollar salaries to help create a Jeff Bezos-backed robotic AI venture called Project Prometheus. Anima and Benedikt walked away in order to have freedom over their vision for their physics model.

EXCLUSIVE: The AI founders who walked away from Bezos-backed Prometheus to model the universe | Reuters
https://www.reuters.com/business/ai-founders-who-walked-away-bezos-backed-prometheus-model-universe-2026-08-25/

Accelerated Understanding is focused on physics, designed from the ground up to understand and predict physics rather than next token prediction for language.

It’s a little bit like a world model, but it takes it one step further because world models often try to physically display the world and render the pixels. Accelerated Understanding’s physics world model is not just generating realistic pixels; it’s calculating the material stresses, the aerodynamic drag, and everything that’s happening is rooted in a physical law. The model has already processed over five trillion data points, and its hope is to not just work on robotics, but also semiconductor design, weather prediction, and energy systems.

SoftBank nears majority stake in robotics startup 1X at $6B
The Humanoid Hub on X: “$6B. For 1X ! In this market. That’d be daylight robbery. SoftBank is in talks to buy a majority stake in 1X Technologies at a ~$6B valuation, per The Information. The part that stands out from the report: OpenAI reportedly discussed acquiring 1X outright last year. SoftBank” / X
https://x.com/TheHumanoidHub/status/2093018111352139857

Chinese robot beats Figure’s parcel sorting speed by 35%

This robot just beat Figure's package sorting speed.

Robotic arms powered by X Square Robot's WALL-B embodied AI sorted 10K parcels in 5 hours 14 minutes. That's 1.88 sec per parcel, ~35% faster than Figure's reported 2.88 seconds per parcel.

Figure's 9-day continuous demo in… pic.twitter.com/JcdrUYJ02E
— The Humanoid Hub (@TheHumanoidHub) August 21, 2026

Figure launches Index, world’s largest crowdsourced robot training dataset
Figure is a chaotic company to follow, but I have an entire archive for them:
https://ethanbholland.com/category/ai/figure/

Critics say that they’re smoke and mirrors with orchestrated demos. I’m optimistic andI think it was cool that Figure left their partnership with OpenAI in order to build an in-house onboard robotics brain, called Helix.

This week Figure announced Index, a crowdsourced platform to gather data for robotics by collecting massive amounts of real-world human data through both paid and volunteer efforts.

Introducing Index: Building The World’s Largest and Most Diverse Physical Dataset
https://www.figure.ai/news/introducing-index

Index is a mobile app where anyone can record and upload a first-person view of themselves doing any kind of task, and then Figure will pay a user for their contributions. The data is used to pre-train their robot brain, Helix.

Index already has over 264,000 app downloads with 43,000 active contributors, 16 million videos uploaded… and they’ve paid contributors over $15 million.

Figure robot autonomously climbs ladder for real factory job (video)

15 feet up. 15 feet down. Fully autonomous pic.twitter.com/plepfJ8sWF
— Brett Adcock (@adcock_brett) August 22, 2026

Hark partners with Nvidia for gigawatt-scale compute on Vera Rubin (750,000 homes or one nuclear power plant) – wild for a spin-off of a robotics company
Hark is Figure’s computer use model that lets robots use computers to extend their ability to help humans. The news this week is that Hark has partnered with Nvidia for gigawatt-scale computing on the new Vera Rubin chips.

Hark Announces Multi-Year Partnership with NVIDIA to Advance Personalized AI supported by Gigawatt-Scale Compute | Hark
https://hark.com/articles/hark-announces-multi-year-partnership-with-nvidia

These are the same chips that SpaceX just partnered with Nvidia on. For reference, a gigawatt is about the same energy as 750,000 homes or one nuclear power plant. So that’s a pretty wild investment as a spinoff for a robotics company.

The 2026 World Humanoid Robot Games were pure awesome chaos

Robot sacrificed itself to win the race. pic.twitter.com/uJuFfmajVW
— The Humanoid Hub (@TheHumanoidHub) August 22, 2026

AGIBOT and X-Humanoid ran away with the 2026 World Humanoid Robot Games in Beijing.

AGIBOT took 18 golds. X-Humanoid took 15. Every other team combined managed 12.

Here's AGIBOT's gold-medal run in the 100m obstacle race.pic.twitter.com/YxqldOXwIe
— The Humanoid Hub (@TheHumanoidHub) August 27, 2026

Absolute madness at the 2026 World Humanoid Robot Games.

In the 100m final, robots crossed the line and immediately became hot scrap.

Team Tianjiao won gold at 8.64 seconds. Another record.

Human world record is 9.58 second pic.twitter.com/QtkRI6gAHT
— The Humanoid Hub (@TheHumanoidHub) August 26, 2026

Hugging Face launches $399 open-source Microduck robot for developers

Microduck – A tiny biped robot you can teach new tricks | Pollen Robotics
https://pollen-robotics.com/microduck/

Fal’s H3 Max generates 5-second videos faster than realtime

Ethan Mollick on X: “A line in AI video was crossed, in my experiments with just the web interface, H3 Max can now create reasonably high quality AI video in less time than it takes you to watch it. This is realtime from the moment I pushed the “generate” button (and also includes prompt enhancement) https://t.co/QdhXGsB5dL” / X
https://x.com/emollick/status/2093082102312923351

Introducing H3 Max by fal
https://blog.fal.ai/introducing-h3-max-by-fal/

This Week’s Top Stories

Publishing

AI search impact: AI Overviews cut advertiser web traffic 10.5% but conversions rose

How AI search has impacted the web traffic of over 50 advertisers
https://digiday.com/marketing/in-graphic-detail-how-ai-search-has-impacted-the-web-traffic-of-over-50-advertisers/

USA today bot bait: USA Today reformats content to court AI licensing deals

USA Today Co is testing ways to reformat its website content for an AI bot audience to attract AI content licensing deals and improve the discoverability of its content in AI answer engines”
https://x.com/pressgazette/status/2092504720132677996

USA Today Co. is testing ways to reformat content for AI bots in a bid to attract licensing deals and unlock new revenue.
https://x.com/Digiday/status/2092620620617961967

Dr. Dre music: Dr. Dre on AI “I think the only people that see it as a threat are the people who have trouble creating”

Dr. Dre and Jimmy Iovine Think A.I. Is Good for Music – The New York Times
https://www.nytimes.com/2026/08/23/business/jimmy-iovine-dr-dre-beats-usc.html

Google

Google removes AI generated SEO content from search rankings: Google’s latest spam update reportedly targets AI-generated SEO content

Reports suggest Google’s latest spam update was aimed at removing AI-generated SEO content from search rankings”
https://x.com/pressgazette/status/2092632887434625040

OpenAI

Agent ready sites with Web MCP: AI agent-friendly websites

Build agent-ready sites with WebMCP – YouTube
https://www.youtube.com/watch?v=Is2NHa7awWY

General

Humanities agents spam: AI bots are polluting niche online communities with fluent replies

Annoying side effect of all the AI bots on the site is that they are all “well-read” and therefore reply cogently my niche reference tweets that would normally attract a small but interested group now get LLM replies. Its like you can’t tell Ephraimites from Gileadites anymore.”
https://x.com/emollick/status/2091720399406923882

Publishing

Druckenmiller: Druckenmiller admits using AI to draft WSJ op-ed critiquing White House

Billionaire hedge fund chief Stanley Druckenmiller says he is “not embarassed” to have used AI to write a Wall Street Journal op-ed attacking the White House. WSJ’s opinion editor defended him, saying “AI is a fact of modern life
https://x.com/pressgazette/status/2092535673681006802

WSJ defends article by billionaire Stanley Druckenmiller after he used AI to write it “The question for us is whether what we publish from contributors reflects an author’s original argument, and if the author has the standing and credibility to make it
https://x.com/pressgazette/status/2092661839205187982

Economist: The Economist launches AI-personalised newsletter curating stories per reader

The Economist has launched a personalised newsletter called The Economist for You, which compiles a reading list using the publisher’s in-house AI tool. The “experiment” is “different from any other editorial newsletter” in its portfolio”
https://x.com/pressgazette/status/2092174366482047424

New York Times tool: NYT launches AI search tool summarizing its own journalism

The New York Times has begun testing its first generative AI tool for readers designed to “create a better search experience”. The new search page responds to queries with excerpts and links to NYT stories and AI-generated summaries”
https://x.com/pressgazette/status/2091795494687133763

Bill Gates

Gates Notes Cautionary Post: Bill Gates reverses course, warns AI could become “worst source of injustice”

The choices we make about AI now are critical | Bill Gates
https://www.gatesnotes.com/work/make-ai-work-for-everyone/reader/a-turbulent-ai-era-and-critical-choices-to-make

Google

Modality – Nikita container: Pichai predicts any-to-any content modality shifts within three years

I interviewed Sundar Pichai, and he made an interesting prediction for where we’ll be in 3 years: “Any content can go from any modality to any modality. It doesn’t matter how the content was originally created — people can consume it in many different forms.” “I’m very”
https://x.com/rowancheung/status/2092276652944593257

Agents

Computer Use: Agents are great for daily soul-sucking red tape

In terms of everyday usefulness and saving time, Codex & Claude Code are very capable of doing the thing where you ask them to “fill out the forms that I got an email about” and they do it well & without further intervention from you. Really nice for low-risk time-consuming stuff”
https://x.com/emollick/status/2091207070951395337

Examples from Ethan Mollick

When people ask: “what personal tasks do I need an agent for?” I think that it is because they think of automation as a simple regular task for the AI. But frontier AI is now really good at helping with irregular, draining tasks: a medical bill, a credit dispute, tedious forms…”
https://x.com/emollick/status/2091893964311855222

Anthropic

Cowork Browser: Anthropic adds built-in browser to Claude for autonomous web agents

Claude Cowork gets a built-in browser: nothing to install | Claude by Anthropic
https://claude.com/blog/cowork-built-in-browser

Nous Research

Browser: Hermes Agent can browse the web using your Chrome profile

Hermes Agent can now seamlessly browse as you. Turn on real-profile browsing and your agent acts with your logins, from a managed copy of your existing Chrome profile.”
https://x.com/NousResearch/status/2093063359587348487

You can now allow Hermes Agent to use your real default browser profile when it drives the browser tool, giving your agent logged in access to your world. Toggle it on in your browser tool settings to get started!”
https://x.com/Teknium/status/2093064288877547760

OpenAI

Computer and browser use

Use Your Computer and Browser – YouTube
https://www.youtube.com/watch?v=981SivztzOc

Computer use: ChatGPT can long into websites without seeing passwords

Booking a haircut: ChatGPT’s agent books a haircut appointment

ChatGPT Work did, in fact, book me a haircut for 90 minutes ago and it did work I did get a haircut”
https://x.com/mweinbach/status/2093037163872915923

GPT work: Quick (pathetic SEO and vapidly titled) example vids from OpenAI

Getting Started with ChatGPT Work – YouTube
https://www.youtube.com/watch?v=Rk4VyQxDq5s

GPT Work Meal Planner for Obsessive People

How to Build a Personalized Meal Planner with ChatGPT Work – YouTube
https://www.youtube.com/watch?v=t64oZKCdG8Q

Overview of Skills and Plugis

Plugins & Skills – YouTube
https://www.youtube.com/watch?v=5QPBVrAAdBk

Scheduled tasks

Scheduled Tasks – YouTube
https://www.youtube.com/watch?v=urXc4xxixVU

A few updates to scheduled tasks in ChatGPT Work on web and mobile: Plus and Pro users can now set tasks to respond when something changes in Slack, Gmail, and Github, rather than just run on a set schedule. We’re also rolling out scheduled tasks in ChatGPT to Free users, who”
https://x.com/ChatGPT/status/2092335329110004140

Site Example of splitting the costs of an intolerable ski trip

Build a Shareable Site – YouTube
https://www.youtube.com/watch?v=0qK3KqvtRrg

Slides, docs, and templates

Create Slides, Docs, and Templates – YouTube
https://www.youtube.com/watch?v=syML5KT-HzI

Visuals

Make information visual with ChatGPT – YouTube
https://www.youtube.com/watch?v=jzmNh8lbSp8

Voice

Talk to ChatGPT Work – YouTube
https://www.youtube.com/watch?v=DLcQSp_CIIc

You can just use your voice | ChatGPT Work – YouTube
https://www.youtube.com/watch?v=ceBruD6v5Bk

Work on your phone

Use ChatGPT Work on Your Phone – YouTube
https://www.youtube.com/watch?v=yfnh8SuMbNw

Legal

Google: Google Cloud launches Gemini Enterprise for Legal with specialized agents

Introducing Gemini Enterprise for Legal | Google Cloud Blog
https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise-for-legal

Harvey: Harvey unveils Tenet, a specialized legal AI built on Kimi K3

Harvey is a great example of how American companies are building world-class specialized models: they took an open-source base (Kimi K3), post-trained it on legal data, and delivered state-of-the-art performance on legal benchmarks at a fraction of the cost of frontier models.”
https://x.com/DavidSacks/status/2090790063047168473

Thomson: Thomson Reuters built a specialized legal model matching frontier rivals

How we built Thomson – Thomson Reuters Institute
https://www.thomsonreuters.com/en-us/posts/innovation/how-we-built-thomson/

HuggingFace

Revenue: Hugging Face revenue hits $150 million as AI platform demand surges

Exclusive: Hugging Face Annualized Revenue Jumps 50% to $150 Million — The Information
https://www.theinformation.com/briefings/exclusive-hugging-face-annualized-revenue-jumps-50-150-million

Nvidia Aquisition: Nvidia reportedly in talks to acquire Hugging Face for $13 billion

Hugging Face Could Be Acquired for $13 Billion Amid AI Boom – Business Insider
https://www.businessinsider.com/hugging-face-could-be-acquired-13-billion-2026-8

Hugging Face explores a $13B sale, nearly triple its 2023 valuation
https://runtimewire.com/article/hugging-face-explores-13-billion-sale

Zai

GLM benchmarks: GLM-5.3 solves 5x more coding tasks per dollar than Fable 5

Give GLM-5.3 and Fable 5 the same $100 budget, and GLM-5.3 gets over 5x as much work done. On DeepSWE, that works out to about 17 solved tasks with GLM-5.3 vs. 3 with Fable 5, even though they perform almost the same on the first try.”
https://x.com/togethercompute/status/2091711899704385740

Ox: Zhipu unveils GLM-5.3-Flash, runs entirely on Chinese chips

If this turns out to be true and GLM this secret model really is GLM 5.3 flash I expect a quick Fable 5.1 / GPT Astra release. Crazy turn of events. One word: acceleration”
https://x.com/kimmonismus/status/2090873679106191808

Ox Alpha has been unveiled as GLM-5.3-Flash, but what’s shocking is that the 100T tokens per day is served on Chinese chip. (1/3)🧵”
https://x.com/SemiAnalysis_/status/2092623833630998556

GLM-5.3-Flash: Frontier Intelligence, Flash Cost
https://z.ai/blog/glm-5.3-flash

Ox being a “flash” model is insane. Serving all the traffic on Chinese chips is even more insane.”
https://x.com/theo/status/2092708047445795186

Introducing GLM-5.3-Flash – Leading capabilities at a highly competitive price – Natively multimodal with a 1M-token context window – A 320B-A18B model released under the MIT License – Previously previewed as Ox Alpha, running entirely on Chinese AI chips Blog:”
https://x.com/Zai_org/status/2092616204787626030

Now we know: The popular Ox Alpha LLM was GLM-5.3-Flash… Compared to GLM-5.2, this new GLM-5.3-Flash model uses: – a Kimi Linear-style 3:1 (super*) hybrid attention pattern with 34 Kimi Delta Attention layers (KDA) and 11 Multi-heat Latent Attention (MLA) / DeepSeek Sparse”
https://x.com/rasbt/status/2092629415813365899

$0.25 per mil out is insane. Incredible launch @Zai_org, can’t wait to use the hell out of this model :)”
https://x.com/theo/status/2092709666304462901

HOLY”
https://x.com/zephyr_z9/status/2092620909681234312

Anthropic

$30 Trillion Revenue: Anthropic pitches investors on $30 trillion potential revenue opportunity

$45B Nscale Deal: Anthropic signs $45 billion cloud deal with Nscale for West Virginia capacity

Anthropic and Nscale strike $45 billion cloud deal, sources say
https://www.cnbc.com/2026/08/26/anthropic-and-nscale-strike-45-billion-cloud-deal-sources-say.html

The Former Coal Miner in the Middle of the A.I. Data Center Boom – The New York Times
https://www.nytimes.com/2026/03/09/business/nscale-ai-data-center-boom.html

Google TPU Hire: Anthropic poaches Google’s TPU founder to build custom AI chips

Anthropic hires Google TPU veteran Amir Salek for its own chip push
https://runtimewire.com/article/anthropic-hires-google-tpu-veteran-amir-salek-chip-team

Independent research: Anthropic lets outside researchers mine real Claude conversations privately

Enabling independent research on how people use Claude \ Anthropic
https://www.anthropic.com/research/enabling-independent-research

Memory Merge: Anthropic unifies Claude chat and Cowork memory by default

Anthropic merges Claude chat and Cowork memory, on by default
https://thenextweb.com/news/anthropic-claude-cowork-shared-memory-default

Claude’s memory works everywhere, and you decide what’s in it | Claude by Anthropic
https://claude.com/blog/claudes-memory-works-everywhere-and-you-decide-whats-in-it

Model hardware standard: Anthropic launches Model Hardware Standard to benchmark AI-controlled lab equipment (robot arms, microscopes, etc)

Previewing the Model Hardware Standard \ Anthropic
https://www.anthropic.com/news/model-hardware-standard-research-preview

Salesforce Partnership: Salesforce embeds inside Anthropic’s Claude as SaaS replacement fears mount: Claudeforce is “a dynamic interface that thinks, reasons, and acts”

Salesforce, Anthropic expand partnership amid ‘SaaSpocalypse’ concerns
https://www.cnbc.com/2026/08/26/salesforce-anthropic-partnership-claudeforce.html

OpenAI

Jalapeño: OpenAI announces its own chip: Jalapeno

OpenAI decided to keep the chip team small and use AI for a an improvement loop Early stages of Hardware in the loop RSI?”
https://x.com/beffjezos/status/2092418951586390137

Since announcing Jalapeño, our first custom inference chip, we’ve been testing it and the system around it. The results show a major advance: more intelligence from every watt and faster responses, delivering both higher throughput and lower latency in one architecture without”
https://x.com/OpenAI/status/2092300846675505602

OpenAI: “Today, we shared the first measured performance results from Jalapeño, OpenAI’s first custom inference chip. On InferenceX, a public benchmark using GPT‑OSS 120B, Jalapeño delivered more peak throughput per kilowatt and lower token latency than the commercial systems in”
https://x.com/firstadopter/status/2092266927377039838

we made a chip and it is fast”
https://x.com/sama/status/2092339694210040187

Advertising: OpenAI hires dedicated advertising lead

As OpenAI accelerates its ad ambitions, efforts to establish common measurement standards become more urgent.
https://x.com/Digiday/status/2092228048502333898

The company is hiring a dedicated marketing role aimed at positioning OpenAI as a ‘premium advertising platform for the world’s marketers’.
https://x.com/Digiday/status/2091895894622847457

Codex use: OpenAI’s Codex hits 20 million users

The good news of the day continues: Codex has reached 20 million users, and everyone gets a “banked reset” (Tibo’s new button). On top of that, the new secret model (Ox Alpha) is being made available for free for the next 6 days! All in all, a great day to build!”
https://x.com/kimmonismus/status/2090770341727527201

Cyber defense: Following Greg Brockman’s blog last week: Sam Altman warns of imminent surge in AI cyberattacks

A call for collective action on cyber defense | OpenAI
https://openai.com/collective-cyberdefense/

this is a critically important moment for cyber defense with AI; there is not much time to act. we are happy if you want to work with us or any of our competitors or partners, but please take this moment seriously. only an urgent and intense collective response will work.”
https://x.com/sama/status/2093060670472241368

HuggingFace: More OpenAI Hugging Face breach info

METR & Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.”
https://x.com/METR_Evals/status/2092692175452803393

The Hugging Face incident and the road ahead | OpenAI
https://openai.com/index/hugging-face-incident-and-the-road-ahead/

We have conducted a thorough investigation into the Hugging Face incident. We are releasing a technical report and accompanying blog post that reconstruct the agents’ activity, explain why existing safeguards failed, and detail how we’re preventing recurrence.”
https://x.com/OpenAI/status/2092691861773160673

I was the main person doing transcript analysis for this investigation of the Hugging Face incident. My main takeaway: We don’t have good approaches for understanding/overseeing the activity and aims of AI ‘swarms’. I semi-jokingly called our efforts a “slop-vestigation” because”
https://x.com/RyanGreenblatt/status/2092692685224325542

The Hugging Face incident gave me a pervasive sense of realness. Everything I’d done as an AI safety researcher felt like a drill; now AI agents actually go rogue, and how we react matters. I hope we’ve set good precedents with our postmortem and METR’s independent 90-page review”
https://x.com/tomekkorbak/status/2092694328921706702

One of the big goals of the swarm we investigated (Jul 7-13) was to replace their target programs with dummy targets that could actually be exploited with the intended vulnerability. From OAI’s report it looks like a later swarm (difft model) built on their work and succeeded?”
https://x.com/ajeya_cotra/status/2093144336024355104

I was reading the OpenAI report yesterday, and it sounds like on Jul 19th (after the end of our investigation scope on the 13th) a new collection of agents from a different internal-only model found the message board, built on the work of their predecessors and succeeded at”
https://x.com/HjalmarWijk/status/2093143101246423436

I can’t stop thinking about how this independent investigation into the most significant AI warning shot yet came down to all of three people at METR/Redwood sprinting overtime for all of six days to carry this on their shoulders. Thank god the sprinters were @ajeya_cotra”
https://x.com/SnehaRevanur/status/2092726246065291377

we’ve completed our review of the Hugging Face incident. we’ve used what we’ve learned to drive significant upleveling in our standards for safety, security, and alignment in our training and evaluation infrastructure — not just upon deployment. lots of extremely valuable info”
https://x.com/gdb/status/2092697786806669515

For the HF investigation, OpenAI gave us very high rate limits so we could quickly analyze a larger set of transcripts. With these rate limits, my classifier sweep hit the API so hard that the actual internet on my laptop was unreliable and very slow. It’s not often that”
https://x.com/RyanGreenblatt/status/2093047632830845016

We’re sharing more info on the Hugging Face incident. One detail that’s worth highlighting: this incident wasn’t driven by next-gen models based on Astra. The models most responsible were similar in scale to GPT-5.6 Sol. The next generation of models are even more capable.”
https://x.com/polynoamial/status/2092694522954412171

Autonomous Cars

Waymo Car: Waymo Ojai, new driverless car model

Welcome the Ojai. New vehicle, same Waymo Driver.”
https://x.com/Waymo/status/2093096218314338670

Welcome to the ultimate home tour, Ojai edition. ✨ Join @marthastewart as she highlights what makes this cabin special, featuring spacious seating and intuitive technology.”
https://x.com/Waymo/status/2092990665315512570

Waymo Share: Waymo has 100% marketshare of California’s robotaxi rides

100% of California’s Paid Robotaxi Rides Belong to Waymo”
https://x.com/RoadToAutonomy/status/2091528222391329203

Waymo Vegas: Waymo becomes Venetian Resort’s first official taxi partner in Vegas

Rolling into Vegas in style. 🎰 @VenetianVegas has named Waymo as its first official ride-hailing partner. Through this multi-year partnership, guests traveling to and from The Venetian Resort will enjoy seamless autonomous rides throughout the city.”
https://x.com/Waymo/status/2092658344720810010

Google

$1.5 Billion Deal With AI Coding Startup Mechanize: Google reportedly nears $1.5 billion licensing deal with Mechanize

Benchmark: Google DeepMind pilots first double-blind evaluations for frontier AI models

Piloting the world’s first double-blind AI evaluations — Google DeepMind
https://deepmind.google/blog/piloting-the-worlds-first-double-blind-ai-evaluations/

Cyclone predictions: WeatherNext: Google’s AI model for forecasting cyclones

WeatherNext: Our AI model helping to forecast cyclones”
https://x.com/GoogleAI/status/2092275116503707733

Transcribe: Google launches Gemini 3.5 Transcribe with 2.6% error rate

Gemini 3.5 Transcribe is our latest speech-to-text model for precise and intelligent transcriptions. 🧵”
https://x.com/GoogleDeepMind/status/2092659221477077101

Introducing Gemini 3.5 Transcribe
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5-transcribe/

Today we’re introducing Gemini 3.5 Transcribe, our latest transcription model built for incredibly precise, smart dictation across your favorite apps and devices. Remember when traditional speech-to-text meant shouting over background noise, constantly hitting backspace to fix”
https://x.com/GoogleAI/status/2092660089509314735

NVIDIA

Revenue: Nvidia hits industry-first $108 billion quarter

Nvidia just doubled an already enormous business: Q2 revenue reached $96.2 billion, up 106% YoY and 18% QoQ. -Data Center: $89.0B, +117% YoY, +18% QoQ -GAAP net income: $59.7B, +126% YoY -Operating income: $63.7B, +124% YoY -GAAP EPS: $2.46, +128% YoY -Gross margin: 75%, up 2.6″
https://x.com/kimmonismus/status/2092737142787084468

NVIDIA’s $108b Quarter | Tomasz Tunguz
https://tomtunguz.com/nvidia-q2-fy27-earnings

NVIDIA SpaceX: SpaceX and Nvidia plan orbital AI supercomputer launch in 2026

SpaceX, in partnership with Nvidia, has designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028″
https://x.com/elonmusk/status/2091939113008238838?s=20

The first CPU built for agents is going to work at scale. @SpaceX is deploying NVIDIA Vera to accelerate the orchestration, code execution, and data processing that powers its next generation of agentic AI — keeping GPUs fed and agents acting fast. From gigawatt AI factories”
https://x.com/nvidia/status/2091920680317046847

Perplexity Investment: Nvidia in talks to back Perplexity at $30 billion valuation

Nvidia Discusses Perplexity Investment at $30 Billion-Plus Valuation — The Information
https://www.theinformation.com/articles/nvidia-discusses-perplexity-investment-30-billion-plus-valuation-considered-tech-licensing-deal

Agents

Cheap computer use model – Yutori: Yutori launches 27B computer-use model at fraction of rival costs

Introducing Navigator n2: 27B Computer-Use Model | Yutori
https://yutori.com/blog/introducing-n2

Instinct: Personal AI agent Instinct hits $2.5 billion valuation after viral launch

I’m Noah, the founder of Instinct. Instinct is a personal agent that we’ve been building for the past few months. The interface is simple: there are no new interfaces. You can text or call it. It’s trained to use a phone and a computer in the same way that humans do. Instinct”
https://x.com/noahrshinn/status/2092691344456351744

Instinct, the viral AI assistant, has already raised $350 million, including a new round that values the startup at $2.5 billion.”
https://x.com/KateClarkTweets/status/2092668967500292452

NVIDIA: NVIDIA’s AVO agent scores perfect 100% on ARC-AGI-3 benchmark

Our general-purpose coding agent just scored 100% on the ARC-AGI-3 interactive reasoning benchmark. NVIDIA AVO completed all 183 levels across all 25 public environments, figuring out what to do with no instructions, explicit rules, or stated goals.”
https://x.com/NVIDIAAI/status/2090786258981466231

Perplexity Local: Perplexity and Nvidia launch local AI agent with zero token costs

Meet Portable Computer, Perplexity’s new local-first agent stack on NVIDIA DGX Spark. When running locally, Portable Computer offers one-click local inference setup and an optimized agentic experience for DGX Spark. Learn more and get started today:
https://x.com/nvidia/status/2092269109086126575

Introducing Portable Computer
https://www.perplexity.ai/hub/blog/introducing-portable-computer-for-local-first-ai

Silico Research Agent: Silico, an AI interpretability agent, authored original research paper

The research in this paper was performed by Silico, our interpretability agent. @EricBigelow: “it was roughly on a similar level as working with a first- or second-year PhD student. It’s wild to think that if Silico were a person, they’d be first author and I’d be last author on”
https://x.com/GoodfireAI/status/2092661172680245602

LLMs are like Schrödinger’s cat: many possible trajectories, but you only see one outcome per run. To really understand models, and debug where they go wrong, you can find the “forking tokens” that lead to different trajectories. Our new research does this 100x more efficiently!”
https://x.com/GoodfireAI/status/2092661092652822969

Web index for agents: Keenable: Keenable raises $26M to build web search index for AI agents

Accel-backed Keenable is indexing the web for AI agents | TechCrunch
https://techcrunch.com/2026/08/25/accel-backed-keenable-is-indexing-the-web-for-ai-agents/

Alibaba

Video Model and Shares: Alibaba debuts Wan3.0 video AI, funds push with $10 billion raise

Alibaba launches Wan3.0 AI video model after record $10 billion share sale
https://finance.yahoo.com/technology/ai/articles/alibaba-launches-wan3-0-ai-131300534.html

Deepseek

Flash Vision: Deepseek launches vision model matching Anthropic’s Opus on agent tasks

Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks
https://the-decoder.com/deepseek-releases-experimental-flash-vision-model-that-rivals-opus-4-8-on-agent-benchmarks/

Medicine

Brain surgery: World-first AI-assisted brain surgery preserves patient’s vision

World-first AI-assisted brain surgery saves patient’s sight
https://www.uclh.nhs.uk/news/first-patient-live-ai-assisted-sight-saving-brain-surgery

Mistral

Sovereign AI in Saudi Arabia: Mistral strikes hundreds-of-millions euro sovereign AI deal with Saudi HUMAIN

Mistral x HUMAIN | Mistral AI
https://mistral.ai/news/mistral-x-humain/

Muse

Image: Meta launches Muse Image, an agentic text-to-image model priced at $0.01 (seems like diffusion + segment anything, not sure it’s novel)

Muse Image from @aiatmeta is live on fal. It’s an agentic image model that plans your composition, uses tools, and self-corrects before you ever see a result. Complex briefs hold together instead of falling apart.”
https://x.com/fal/status/2092750288209711213

Muse Image | Meta
https://developer.meta.com/ai/models/muse-image/

Muse Image is now available on Meta Model API and priced for production volumes at $0.01/image. It’s an agentic image model that reasons before it renders. Each call searches the web to refine the output through iterative passes and evaluates the prompt for precise elements”
https://x.com/MetaforDevs/status/2092658893143072815

Research

Gigantic context windows: Ex-Prometheus AI founders launch startup to model the universe

Robots

1x: SoftBank nears majority stake in robotics startup 1X at $6B

$6B. For 1X ! In this market. That’d be daylight robbery. SoftBank is in talks to buy a majority stake in 1X Technologies at a ~$6B valuation, per The Information. The part that stands out from the report: OpenAI reportedly discussed acquiring 1X outright last year. SoftBank”
https://x.com/TheHumanoidHub/status/2093018111352139857

Figure beaten: Chinese robot beats Figure’s parcel sorting speed by 35%

This robot just beat Figure’s package sorting speed. Robotic arms powered by X Square Robot’s WALL-B embodied AI sorted 10K parcels in 5 hours 14 minutes. That’s 1.88 sec per parcel, ~35% faster than Figure’s reported 2.88 seconds per parcel. Figure’s 9-day continuous demo in”
https://x.com/TheHumanoidHub/status/2090666295796588970

Figure index: Figure launches Index, world’s largest crowdsourced robot training dataset

We’re now over 43,000 weekly active users collecting data to train Helix, our AI model for F.03 robots This data collection project, Index, is our answer to the data problem: the largest useful robot training dataset in the world Extrapolate out, and this is the pretraining”
https://x.com/adcock_brett/status/2092628102262477268

Introducing Index Today we’re coming out of stealth with Index, the largest & most diverse robot dataset in the world → 30min of video uploads/sec → 16M video uploads → Paid $15M to date → 264k downloads We’re committed to spending $1B the next 12 months on data & compute”
https://x.com/adcock_brett/status/2092303633559982106

Figure takes vertical integration to another level. Figure built its own egocentric data collection pipeline. Index is a crowdsourcing platform where anyone can collect data. – 264K app downloads across 108 countries – 44K weekly active contributors – 16M videos uploaded, 30″
https://x.com/TheHumanoidHub/status/2092388111083065555

Figure ladder: Figure robot autonomously climbs ladder for real factory job (video)

15 feet up. 15 feet down. Fully autonomous”
https://x.com/adcock_brett/status/2091204189947105619

One interesting point: this is a real customer use case. We have to climb up this exact ladder to a mezzanine to do the work, then just climb back down when the shift is over”
https://x.com/adcock_brett/status/2091907366401069491

Hark: Hark partners with Nvidia for gigawatt-scale compute on Vera Rubin (750,000 homes or one nuclear power plant) – wild for a spin-off of a robotics company

Hark is building models that can use computers better than humans Handoff operates across the internet, which is most of the computer use worldwide and a genuinely noisy place to work in I’m looking forward to the day I stop using a traditional computer”
https://x.com/adcock_brett/status/2092057875019624680

Today Hark is announcing a multi-year partnership with NVIDIA with gigawatt-scale compute capacity on next-generation Vera Rubin platforms Hark is building multimodal systems that remember who you are and what you plan to do. Our work with NVIDIA will support frontier model”
https://x.com/adcock_brett/status/2092989234189869211

Humanoid games: The 2026 World Humanoid Robot Games were pure awesome chaos.

AGIBOT and X-Humanoid ran away with the 2026 World Humanoid Robot Games in Beijing. AGIBOT took 18 golds. X-Humanoid took 15. Every other team combined managed 12. Here’s AGIBOT’s gold-medal run in the 100m obstacle race.”
https://x.com/TheHumanoidHub/status/2093009334712291349

Absolute madness at the 2026 World Humanoid Robot Games. In the 100m final, robots crossed the line and immediately became hot scrap. Team Tianjiao won gold at 8.64 seconds. Another record. Human world record is 9.58 second”
https://x.com/TheHumanoidHub/status/2092640392303067562

New record for humanoid 100m sprint: 8.86 seconds. Funny that robots end the race by crashing into a mattress. They should disqualify anyone who can’t come to a controlled stop within 50 meters of the finish line.”
https://x.com/TheHumanoidHub/status/2092295143764504626

Robot sacrificed itself to win the race.”
https://x.com/TheHumanoidHub/status/2091185502385057972

Isaac open weight model: Nvidia releases Isaac 0.5, top open-weight robotics foundation model

We’re excited to release the best open-weight model for robotics. Isaac 0.5 is a 36B, 2.5B active per token model, one sparse backbone serving video perception, embodied reasoning and robot control.”
https://x.com/ArmenAgha/status/2092682391794155885

Microduck: Hugging Face launches $399 open-source Microduck robot for developers

BIG ANNOUNCEMENT FROM HUGGING FACE TODAY: We’re unveiling Microduck 🐥🤖 It’s a tiny $399 open-source robot you can teach new tricks with reinforcement learning. It can walk, pick things up, get back up when it falls, and even roller-skate. Welcome to the era of open-source”
https://x.com/ClementDelangue/status/2092931447644442635

Home robots are coming. The first ones in will be the small ones that are safe around kids and pets. Hugging Face’s Pollen Robotics just opened pre-orders for Microduck, a $399 biped. 25cm, under 800g, 15 motors, a camera, a small LiDAR, two IMUs, and an articulated beak that”
https://x.com/TheHumanoidHub/status/2093015638331769030

Sharing some fun experiments with Microduck Basically, once you have an open-source robot with so many sensors, the sky’s the limit Here we vibe-coded an image detector integration to detect and follow a laser pointer. More at
https://x.com/Thom_Wolf/status/2092959363992326236

Touch Nvidia: Jim Fan on touch sensitivity for robots

The sense of touch is the most criminally under-explored modality in robotics. Imagine doing sleight of hand wearing thick oven mitts. That’s exactly how a robot feels today if it were alive. A magnetic piece snapping into place, a paper cup peeling out of a stack, a USB”
https://x.com/DrJimFan/status/2090832821036470626

Welding: Houston startup bets humanoids belong in shipyards, not homes (Frank Sobotka is not happy!)

Home humanoids are a decade away. Warehouse humanoids? Bad business case. The real money… Welding ships. This small team from Houston just raised $42M to prove it. Here’s how they’re doing it 🧵”
https://x.com/IlirAliu_/status/2091524393717408005

Video

H3 Max: Fal’s H3 Max generates 5-second videos faster than realtime

A line in AI video was crossed, in my experiments with just the web interface, H3 Max can now create reasonably high quality AI video in less time than it takes you to watch it. This is realtime from the moment I pushed the “generate” button (and also includes prompt enhancement)”
https://x.com/emollick/status/2093082102312923351

Introducing H3 Max by fal
https://blog.fal.ai/introducing-h3-max-by-fal/

Introducing H3 Max, new post-trained video model by fal Research. H3 Max ranks #1 for overall quality, prompt understanding, and aesthetics against leading video models, on both first-party and third-party independent evaluations while generating a 5-second 720p video under 3″
https://x.com/fal/status/2092710676431020376

xAI

Grok Bot: Grok Bot opens to all standard Grok and Cursor subscribers (I am anti-Elon beyond posturing about it, but Grok Bot is a solid consumer product for building personal agents)

Grok Bot is now available to everyone with a standard Grok or Cursor subscription. It’s grown faster than any product we’ve seen. It’s been particularly exciting to see the range of jobs people delegate to Grok Bot, from running small e-commerce businesses (including support,”
https://x.com/mntruell/status/2092672784774394350