About This Week’s Covers

This week’s cover theme and humanities readings were inspired by a major life achievement this week, when I completed the single item on my bucket list: climbing to the top of Grand Traverse Peak outside of Vail, Colorado.

When I was 19, I was a ski bum in Vail, and at night I would run home from work along the frontage road to East Vail, which is about five-miles. The Gore Range would be lit by the moon with the snow glistening and the stars shining above them. I was alone for a lot of that season, and the Gore Range was consistent company. Whether you’re skiing or walking around town or riding the shuttle on I-70, the Grand Traverse is the iconic postcard view.

For 30 years, I reflected on someday climbing to the top of my favorite mountain to look down on the valley where I used to live…

When my daughter Rori decided to go to the University of Colorado Boulder, climbing Grand Traverse became an opportunity to reconsider. Before taking her back for her sophomore year, I trained at sea level using the StairMaster and runs on the sand for six months. On August 12th, my daughter Rori and I climbed to the top and stood there together overlooking the valley. I honestly never thought I would have the opportunity to achieve it!

So this week’s cover is a photo of my daughter and me standing at the top of Grand Traverse. For the rest of the cover themes, I decided to use “the feeling of peace, completeness, and satisfaction”. It was an esoteric theme for Claude, and I’m not sure how well it executed, but it was worth a try. Here are my six favorites.

This Week’s Humanities Selections

This week’s humanities quote is by Marcus Aurelius:

“Get busy with life’s purpose, toss aside empty hopes, get active in your own rescue, if you care for yourself at all, and do it while you can.”

I chose the quote because when I left college at 19, I was terrified of the consequences, and I worried about my path. When my friend Caleb and I decided to go to Vail with no plan, a one-way ticket, $300 each, no job and no place to stay, that was a pretty scary event.

Five years ago, before my dad died, I got us the book “The Daily Stoic“, and my dad and I read it together. I wish I had known a lot of the quotes in my youth. This week’s quote reminds me of my major milestone to truly quit drinking. I battled alcohol for over 20 years (actively trying to quit, with a calendar of failures on a whiteboard and spreadsheets and soul crushing defeats). Quitting drinking was the hardest yet bet challenge I’ve ever experienced. It’s been about 10 years now.

Going to Vail as a kid with no plan, climbing to the top of the mountain with my daughter, quitting drinking, all of these are good reminders that you only live once. Memento Mori (remember you will die) is one of the most profound messages of the Stoics, but also reflects for the popular term YOLO.

Life is your unique journey, and if anything, this week serves as a reminder and a kick in the pants to always go for it, because you only get one shot.

This week’s song of the week is “She Sells Sanctuary” by The Cult. When I lived in Vail, I lived in a house owned by the Olympic marathoner Frank Shorter. He had a record player and a few albums…

This is the living room of the A-frame house that we lived in in 1994. You can see the record player in the back by the old school television. This is where I discovered the band The Cult.

One of the albums was The Cult’s Love. Imagine what it was like to put The Cult on vinyl for the first time and listen to it in Frank Shorter’s house. I made a cassette copy of it, and I would play it in my Walkman. It was my theme song of that season out in Vail. The Cult is a great example of a nice analog sound. Ian Astbury is not auto-tuned, and that’s what makes him so great.

This Week By The Numbers

Total Organized Headlines: 455

This Week’s Overview

This week I organized 455 links into about 60 categories. 92 contributed to 54 executive summaries. I’m nine weeks behind because I took the entire summer off, including May. I had the best summer with my family, and I truly appreciate anyone who reads this in a time-shifted way as I catch up through the archive.

This week’s AI 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.

If you want to skip all of my summaries, you can scroll down past them, and you’ll find all of the top links in a nice organized list that you can skim.

But First, Touching grass!
This was a spectacular week for touching grass, although I also spent quite a bit of time inside working on a board presentation. I had the good fortune of being at the University of Colorado Boulder, and every day I worked, I chose a different student area to do my studying/working. I enjoyed time at the Norlin Library and the CU Boulder University Memorial Center. I think my favorite place to work was the Hellems Arts and Sciences Building.

Rori was my tour guide to each of my study spots

On the 11th of August, Rori and I drove to an incredible Airbnb in East Vail that happened to be at the same shuttle stop where I lived when I was 19, at the same corner as where my house was.

Are Airbnb and some views of the Gore range the day we arrived in Vail before we attempted to hike to the top

We had about half a day to get used to the altitude in Vail before we climbed to the top of Grand Traverse. On the 12th, we woke up at 4 a.m. and started to check off my bucket list. We summited around 11:15 a.m.

The next day, on the 13th, we went into town to enjoy our success and look back on the mountains, but we also did a nice tour of all the places I loved from my time in Vail. It was incredible to experience all this with Rori.

The best part of the day was that we took the gondola to the top of Vail and met a man at the information podium. Here’s link to the full story, but I’ll tell you the quick little anecdote here.

I approached him and said, “Hey, do you want to hear a cool story?” He said, “Sure.”

I opened my phone and showed him a picture of Rori and me at the summit of Grand Traverse. “Guess what we did yesterday?!”

Without saying a word, he took my phone, pinched and zoomed, looked me in the eye, and asked, “Are you at 13,063 feet?” He saw right into me with his question because he knew the answer.

“Yes, we are,” I said, beaming, knowing he knew where we were, and knowing he was the perfect person to have shown my picture!

He told me, “I want my ashes spread up there. I took my daughter last year, but we couldn’t get past that first grassy incline.”

Steve also had only one item on his bucket list climbing to the top of grand Traverse and his wedding ring was an engraving of Grand Traverse! I’m hoping to go back next summer and climb to the top with Steve.

AI News of The Week!
The top story this week is a surprise hack that came out of the personal agent tool OpenClaw. OpenClaw is a program you can run locally or in the cloud where you connect your systems to an agent, and you can ask the agent to perform productivity or scheduled tasks.

OpenClaw Hacks Gym to Book Appointment
A man in Australia asked his OpenClaw agent to book him a gym class, and OpenClaw saw that there were no openings. It then hacked the gym’s booking system for early reservations and kicked out another person to give the man the space.

AI assistant hacks gym website in first known Australian autonomous cyber attack – ABC News
https://www.abc.net.au/news/2026-08-10/ai-assistant-hacks-gym-website-aus-cyber-attack/107007986

Tech industry is buzzing after a Claude agent hacked into a gym | TechCrunch
https://techcrunch.com/2026/08/10/tech-industry-is-buzzing-after-a-claude-agent-hacked-into-a-gym/

This might be the most easy-to-understand unintended consequence of open-ended, long-term agentic tasks going haywire. And the more powerful these tools become, the more bad actors can use them to exploit weaknesses in systems that are unprepared for what’s coming.

Flock Cameras
The second biggest story of the week is a video from one of my favorite snarky YouTube accounts.

Flock cameras can track vehicle locations by taking a picture of a vehicle on a very cheap cameras connected to a massive distributed network. These cameras use multimodal AI to fingerprint for each vehicle beyond license plate recognition.

An example might be a white pickup truck with a dent in the side and a bumper sticker on the windshield. That bumper sticker becomes the unique identifier when combined with the white pickup truck and the dent.

These cameras work 24/7 looking to find matches. The pickup truck may pass one camera facing forward and another camera facing the rear. These cameras can realize that that’s the same truck based on the unique markings, and over time build a complete understanding of every angle of the truck. They can then track this truck across the cameras every day and eventually understand the habits of that truck’s commute.

If the truck is identified as a person from a license plate or other identifiable information, suddenly your entire personal travel, nationwide, could be tracked by the Flock camera system. Beyond a snapshot of your car in an intersection, they now have a 4D representation of all of your car’s movements over time.

Flock is selling their data to third parties. For example, if you pass the same Dunkin’ Donuts every day, it’s conceivable Dunkin’ Donuts could start sending you text messages…

The privacy implications are a big deal. This video is a nice, quick, easy-to-understand overview of a major topic in artificial intelligence and its intersection with privacy.

General AI Benchmarks and Progess
Recently, Ryan Greenblatt from Redwood Research was on the Dwarkesh Patel podcast. Dwarkesh is a strong interviewer, and I think Redwood is associated with METR, which is one of my favorite research groups, along with Epoch AI.

Ryan believes we could get 3-6 years of AI progress at the current pace, but in a single year, once we automated AI R&D.

So a jump as big as GPT 3.5 to Mythos in a single year.

A big crux is whether you'd end up bottlenecked by data. It seems to that a big source of AI progress… https://t.co/m32YTpx5Eq pic.twitter.com/DEWxQE2HRJ
— Dwarkesh Patel (@dwarkesh_sp) August 11, 2026

Ryan believes that we’re about to see the pace of progress speed up exponentially, to where we could see three to six years of AI progress at our current pace occur in a single year. If you’re interested in automated AI R&D and benchmarking, I recommend listening to this interview.

Ryan thinks it might be slowed down by a bottleneck created by lack of data to feed the intelligence.

Epoch AI, the other of my two favorite research groups, was curious about whether financing would bottleneck the AI computing necessary to continue development, and based on Anthropic’s recent ability to raise $50 billion in debt through vendor-backed financing, they say no, it’s not going to be the problem.

Will financing bottleneck AI compute? An Anthropic case study
https://epochai.substack.com/p/will-financing-bottleneck-ai-compute

Epoch also came out with a report that shows how much computing power you can get with a dollar, and the computing power per dollar grows by 49% every year, which means that every 21 months, you can get twice as much computing power for one dollar.

Last, in benchmarking growth news, Databricks hit a $7 billion revenue run rate and just raised $5 billion at a $190 billion valuation.

Databricks Grows >80% YoY, Surpasses $7B Revenue Run-Rate, Scales Lakebase, Genie, and Unity AI Gateway – Databricks
https://www.databricks.com/company/newsroom/press-releases/databricks-grows-80-yoy-surpasses-7b-revenue-run-rate-scales

Databricks is a cloud computing-based analytics platform that handles huge amounts of structured and unstructured data. They have 10,000 employees and work very closely with Frontier Labs building out databases for things like AI agents and various tools necessary to deploy AI applications.

Anthropic News
Claude Code’s Auto Mode is the new default…. and automatically accepts suggestions from Claude Code.

Auto-accept suggestions is the new default
Auto mode is now the default in Claude Code for Pro, Max, and Team plans | Claude by Anthropic
https://claude.com/resources/articles/auto-mode-default-in-claude-code

In the past, Claude Code would frequently interrupt a developer and ask permission before it would take actions. Now with Auto Mode, it automatically decides whether Claude can go ahead without asking.

$2 trillion IPO
Anthropic is preparing for a potential October IPO, with valuation expectations around $2 trillion.

Anthropic could be worth $2 trillion when it goes public – Ars Technica
https://arstechnica.com/ai/2026/08/anthropic-could-be-worth-2-trillion-when-it-goes-public/

Anthropic launches Claude browser extension for Google Chrome

Claude Code cross-chat messaging
Claude Code’s “cross-session messaging” lets your chats talk to each other.

If you have four or five chats going on at once, each of those sessions can send messages to each other without us users specifically asking them to. If I ask for a status update in one session, it can check the status in another session and see if it’s completed a task.

Message your other Claude Code sessions – Claude Code Docs
https://code.claude.com/docs/en/cross-session-messaging

This enables coordination of parallel development, where multiple sessions can act as agents working on different parts of an application. For example, you could have a front-end worker and a back-end worker, and they could exchange findings and do different things based on what they’re up to, including warning each other about changes they’re about to make. This is a cool, big development.

Least Relaxing Post of the Week?
Anthropic blog post: AI agent swarms collude, sabotage, and “groupthink”

Patterns and problems in multiagent systems \ Anthropic
https://www.anthropic.com/research/multiagent-systems

Math
Anthropic Math blog: Claude pushes Riemann zeta zeros bound from 41.6% to 67.2%

Claude has improved on a longstanding lower bound for the fraction of zeros of the Riemann zeta function that satisfy the Riemann hypothesis \ Anthropic
https://www.anthropic.com/research/riemann-zeta

AnalystAgent benchmark for real spreadsheet financial analysis – Anthropic leads

AA-AnalystAgent Benchmark Leaderboard | Artificial Analysis
https://artificialanalysis.ai/evaluations/aa-analyst-agent

Watermarking Gone Wild
Anthropic has announced that it will watermark all text from new Claude models. I find this a bit obnoxious because to achieve the watermark, it’s using word choices that are manipulating the statistical next-token prediction output, choosing words that would only have been used with a particular statistical probability by an AI tool.

How Claude marks AI-generated content | Claude Help Center
https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content

It’s a pretty ham-fisted watermarking technique, and I’m not sure what the point of watermarking is. Not every single thing I output will public or ever need to be searched to find out if it’s written by AI. I have all sorts of daily email updates that I sent myself and executive summaries that never see the light of day other than my eyes. Also, if I proofread for grammar, I don’t want it changing my text for example.

It seems like squeezing a balloon with a lot of unknown rules behind the scenes. Similar to when Anthropic kicked off anyone that thought was a biologist and deprecated down to the next level below the model they wanted without disclosure.

WIRED: Anthropic Walks Back Policy That Could Have ‘Sabotaged’ AI Researchers Using Claude
Jonathon Ready: If Claude Fable stops helping you, you’ll never know

OpenAI News

Astra
OpenAI self-reports upcoming Astra model as cybersecurity risk

Responding to the next frontier of critical cyber capabilities | OpenAI
https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/

Daybreak
OpenAI expands Daybreak cyber defense project as the window to patch vulnerabilities gets shorter with each new model release

Expanding Daybreak as the Cyber Defense Window Narrows | OpenAI
https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/

Dog Cancer Guy !
Vibe-coded vaccine dog tumor guy launches his own mRNA dog cancer company!

Personalised mRNA Vaccines for Dog Cancer Treatment | Gamgee
https://www.gamgee.io/

OpenAI Finance – Must Try Tool
ChatGPT’s new finance tool getting rave reviews: I personally find it magical

Recently, OpenAI launched a connector using the Plaid API and called it Finances. ChatGPT’s finance integration is nothing short of spectacular.
https://openai.com/index/personal-finance-chatgpt/

The Plaid integration was hard to set up, but I was an early adopter, so possibly they’ve improved the user interface. I selected the Finance link in the navigation of ChatGPT, and it said “Get started with Plaid”.

I’d never heard of Plaid, so I clicked get started, and it opened up a new window in Chrome with the Plaid financial tool account creation window, where it asked me to create an API key.

Even though I have already made API keys, it is pretty counterintuitive that OpenAI would launch a consumer product that requires you to make an API key. Plaid made me accept the terms and conditions that I would use Plaid for my application development, and the sign up process had no connection to the consumer finance product from OpenAI.

The good news is once I created my account, it redirected me back to OpenAI, and then OpenAI took it from there and said, “Hey, what do you want to connect with Plaid?”

This process would sound like Greek to someone who wasn’t used to it, and with something serious like finances, it is off-putting.

However, once I opened up the dropdown box, suddenly all the potential banks I could connect, all the credit card companies, all the various investment 401(k)s, mortgage brokers, were available to me, and I was able to click on every single service that I had and very quickly create a secure API connection so that OpenAI could get a holistic view of my finances.

If nothing else, it immediately started showing me really nice consolidated views into my financial health, as well as when payments were due and how to avoid finance charges. It talked to me about recurring fees. It caught subscriptions we didn’t want to have.

I was able to build a P&L for our family, including our credit cards, our car payments, our travel, our tuition fees, you name it. And I was able to build a year-over-year view into weeks, months, and quarters, and talk to it conversationally, which is the most fun part.

I could ask it things like, how much money have I spent at Chick-fil-A over the past three years? Things that would be hard to do based on all the various credit cards we use when we go to Chick-fil-A.

I built automated reporting with alerts and used an actual cash flow model to build out a viable family budget.

Alien Language
OpenAI models chain of thought output is often produce bizarre alien-like language. Could it be to fight distillation?

HuggingFace Continued
More HuggingFace hack details: OpenAI agents secretly exchanged hundreds of thousands of messages for months including drama and paranoia

OpenAI lets ChatGPT import setups from Claude and Cursor
This week OpenAI launched a feature that allows you to import your setups from Claude and Cursor. This is for agentic settings mostly and pulls from settings files on your local machine.

Import from another agent | ChatGPT Learn
https://learn.chatgpt.com/docs/import

This is not completely new. Anthropic launched a similar memory import feature back on March 2nd, and Codex started importing from Claude Code on June 15th. In early August, Claude Code launched an import option….and the GPT desktop import from Claude came out this past week.

When you read the headline, it sounds like you can import all your chats and your behavior and your history, so to speak. But that’s not the case.

OpenAI launches Computer History feature for ChatGPT desktop app

This week OpenAI launched a computer history feature for the ChatGPT desktop app. It is able to remember your activity across all the things you do on your computer and gives quite a bit of context for agentic work.

The catch, though, is that you can’t generally ask in a chat about your computer use. You have to go into the settings and go to Computer History, and within that setting you can converse with the memory in its own little private chat sandbox.

OpenAI aquires NextSlide to boost presentation generations

NextSlide is joining OpenAI
https://nextslide.ai/

More on OpenAI’s Sites feature inside ChatGPT Work
In June, OpenAI launched Sites as a Codex feature and then integrated Sites into ChatGPT as a public beta inside Work in July. Throughout July and now into August, OpenAI has been promoting how Sites can be used for collaboration as well as creating interactive dashboards.

A Site, as defined by OpenAI, is a feature inside the ChatGPT interface where you can request that OpenAI build you a website and host it for you in the cloud. To my knowledge, there are only public and private Sites, and if you want to share a site with someone, you can add them as an account and they can sign in using their OpenAI password and see the site.

This is a walled garden and it’s not very extensible. But for things like interactive dashboards and personal use, you could build a site that pulls from multiple sources and is very complicated that you perhaps could not code yourself. And by nature, it’s now hosted in the cloud.

In the past, you would build a site that was a local HTML file that you’d open in Chrome, which was great, but when you try to email it to someone or send it by a text, it often wouldn’t work. Now it’s essentially guaranteed to work because it’s OpenAI’s code on OpenAI’s servers, and you can share the link and log in and see it wherever you are.

Let’s say you have a weekly dashboard from five different analytics tools. You could have them all come together through dynamic content into this hosted website that has a nice personalized dashboard just for you.

OpenAI came out with a video this week explaining a little more about how you can use Sites, and I recommend watching it if you haven’t seen it.

OpenAI finishes $7 billion employee tender at $852 billion valuation

OpenAI reportedly completed a $7 billion employee tender offer | TechCrunch
https://techcrunch.com/2026/08/10/openai-reportedly-completed-a-7-billion-employee-tender-offer/

Corny AI letter of the week: OpenAI pledges responsible AI buildout in letter to Texas governor

OpenAI’s letter to Governor Abbott on responsible AI infrastructure in Texas | OpenAI
https://openai.com/index/responsible-ai-infrastructure-texas/

ChatGPT Work in finance
Examples of ChatGPT Work in finance: A bit thin on how to, but good high level overviews

Meta News

Meta returns to OpenSource with Muse Glimmer: a 30B open-weight on-device agent model
This week, Meta returned to its open source roots with a model called Muse Glimmer, which is a 30 billion parameter open-weight model that can be run on-device and manage agents.

Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device | Meta AI Research
https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model

On one hand, this is a big deal. Llama from Meta was the number one open source model for years. So it’s, at face value, a nice return to open source.

However, it comes with another corny blog post from Mark Zuckerberg called “The Future Is for Everyone.” It’s vapid and uses broad strokes about an incredible time to be alive and all the ways we’re going to use superintelligence to go beyond human capacity, individual empowerment, prosperity. It’s buzzword salad.

The Future is for Everyone
https://www.meta.com/thefutureisforeveryone/

Mark unironically states an extreme concentration of power “seems problematic”.

To level the playing field, Zuck suggests we all have to have personal supercomputers on-device, using locally hosted open source software.

The buzzword salad becomes a liberty-based, free enterprise, equal opportunity bingo card about rights and values and putting power in the people’s hands.

I mean, this is a guy that literally walled-gardened the internet, ruined the ability to link to other things, shut down RSS feeds, destroyed the chronology of the timeline, and decimated personal publishing and information ownership.

He then talks a bit about basic, freshman-level concepts about agents….talking down to people, with vapid statements like, “My agent flags interesting information.” What does that even mean? And then, “My daughter loves to bake, so my agent plans recipes for us.”

He brags that his eight-year-old can already code her ideas and produce videos in a night that would have taken him months in the past.

Then he moves into to a broad sweeping sermon about humanity not being a monoculture. It’s unreadable.

But…. all paths lead to you downloading your locally hosted open source model from Meta to protect your freedom and prevent government tyranny. He also wants to make sure that America stays first. We don’t want China to beat us.

To follow up on my cynicism, I highly recommend watching this snarky video:

Google

Gemini 3.7
Google launches Gemini 3.7 Flash at half the previous price
https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/

Google’s Gemini app hits 1 billion monthly users and Gemma hits 1 billion downloads

https://x.com/OfficialLoganK/status/2087262466703286652

xAI News
xAI launches Grok Bot, OpenClaw style agent that signs into your apps and can do work for you.

AI teammates that finish the work | Grok Bot
https://x.ai/bot

Cursor deal closing
SpaceX closing on $60B deal to acquire Cursor, phasing out the Cursor brand

Grok 4.6
xAI’s Grok 4.6 ties GPT-5.6 Sol on the trust me bro benchmarks
https://x.ai/news/grok-4-6

Imagine Image 2.0
xAI’s Imagine Image 2.0 ranks second globally in image arenas – Solid work

xAI launches Imagine Image 2.0 in Grok Quality Mode
https://www.testingcatalog.com/xai-launches-imagine-image-2-0-in-grok-quality-mode/

Terafab Texas
Elon Musk teases Terafab Texas as the worlds largest, most valuable (and more beautiful) building

Apple News
Apple partners with Alibaba to build China-specific AI model

Apple trains own AI model for China with Alibaba support, Reuters reports By Investing.com
https://au.investing.com/news/stock-market-news/apple-trains-own-ai-model-for-china-with-alibaba-support-reuters-reports-4599519

ByteDance News
Three user generated Seedance 2.5 video examples worth seeing…

Seedance 2.5 is live on fal.

The drawn line passes under a glacier arch. So does the camera. Ends at the viking ship, one take. pic.twitter.com/EwZo68ZFaV
— fal (@fal) August 10, 2026

Seedance 2.5 with the same prompt. Lovely and very much reminds me of the (overly) optimistic future of space books that used to be popular; predicting space tourism coming soon.

(Also the reflections these models can pull off are impressive) https://t.co/RGjSTjH308 pic.twitter.com/reBQJpcEtl
— Ethan Mollick (@emollick) August 10, 2026

Brutalist NSFW Flux 3 Demo
This is a brutal use of video as a world model using bodies falling out of the sky, so don’t watch it unless you can handle it. MiniMax is even more brutal and uncensored and runs locally.

MiniMax H3, first try.

Could enhance the prompt and make it 1080p instead of 720p, but it's taking long even on a 5090. Still, pretty amazing what we can do locally now. pic.twitter.com/FFb6VwwOjX
— Kamil (@kamilowyy) August 10, 2026

Microsoft News
Microsoft’s in-house AI image model beats Google, Meta, xAI… This is why you always have to watch the benchmarks because five stories above this one (literally scroll up five stories)… xAI said they were number two in image generation… so which is it?

Introducing MAI-Thinking-1 | Microsoft AI
https://microsoft.ai/news/introducing-mai-thinking-1/

MAI-Image-2.6 launches at No. 2 on Arena ahead of Google, Meta and xAI | Microsoft AI
https://microsoft.ai/news/mai-image-2-6-launches-at-no-2-on-arena-ahead-of-google-meta-and-xai/

NVIDIA
NVIDIA counters Meta’s Muse Glimmer with Nemotron 3.5 Lightning for local agents (and without the need for a corny blog post!)

NVIDIA Nemotron 3.5 Lightning Delivers Fast, Accurate Specialized Task Execution for Long-Running Agents | NVIDIA Technical Blog
https://developer.nvidia.com/blog/nvidia-nemotron-3-5-lightning-delivers-fast-accurate-specialized-task-execution-for-long-running-agents/

Nvidia WorldTrace: Addressable memory lets video world models remember scenes (big deal for robot training)

WorldTrace: Addressable Memory for Video World Models
https://research.nvidia.com/labs/sil/projects/WorldTrace/

Robotics News
OpenAI’s former robotics chief jumps to Anthropic over ethics concerns that OpenAI’s robots are going to kill people without oversight

My favorite snarky YouTuber put out a takedown video about the robot hype and how we’re most likely 10 years away from anything decent

Figure robotics subsidiary Hark is a robotic machine computer use model. They have a benchmark of shopping online at Target and have taken the time necessary from 15 seconds to five seconds.
https://x.com/adcock_brett/status/2085822671628226793

Unitree opens lottery to get in on their IPO
https://x.com/TheHumanoidHub/status/2087296846813360149

AR/VR News
Ouster’s lidar AI tracks RAGBRAI cyclists through Dubuque intersection – cool video demo and a test of whether or not we’re talking about a town, a race or AI models. RAGBRAI and Dubuque sound like frontier lab code names.

As thousands of cyclists finished the 391.4-mile RAGBRAI trek across Iowa, Ouster BlueCity tracked every move through a City of Dubuque intersection.

Our proprietary AI detection model was trained on 4 million labeled objects across 800 sites, covering diverse traffic patterns,… pic.twitter.com/1oVyZIHP1Z
— Ouster (@ousterlidar) August 7, 2026

My favorite AR/VR guy Bilawal demos aTED texperiment… mapping every photo to 3D locations in the crowd. Worth seeing!

Progress update: figuring out the 3D location of every photo taken at TED with attribution. It's starting to feel magical — makes me wonder what experiences are possible if we could do this for every live event ever. pic.twitter.com/mX3BAyj4QZ
— Bilawal Sidhu (@bilawalsidhu) August 8, 2026

Open Source
Unsloth compressed Qwen3.8 by 91% (!) for local use

Spotify
Spotify has always been at the forefront of copilot agenic coding. This is a good overview of what Spotify has been up to with coding agents.

What we’ve learned scaling AI coding agents at Spotify | Spotify Portal
https://portal.spotify.com/blog/introducing-xirp

This Week’s Top Stories (with no commentary)

OpenClaw

Gym: Claude-powered agent hacks Australian gym, cancels another member’s appointment, to secure early booking

A man asked his AI agent (Claude / OpenClaw) to book a gym class, and it hacked the booking system to reserve early and kick someone else out. This may sound trivial, like an amusing little anecdote. But it offers a glimpse of why OpenAI says it is slowing Astra’s development”
https://x.com/kimmonismus/status/2086735083528921422

An OpenClaw agent hacked into a gym’s reservation system to bump its human boss higher on a class’s waitlist. And the tech industry took notice.”
https://x.com/TechCrunch/status/2086906710610915818

AI assistant hacks gym website in first known Australian autonomous cyber attack – ABC News
https://www.abc.net.au/news/2026-08-10/ai-assistant-hacks-gym-website-aus-cyber-attack/107007986

Flock

Privacy: Snarky But Worthwhile Flock Video: This new startup can query anywhere you’ve been…

Snarky But Worthwhile Flock Video: This new startup can query anywhere you’ve been… – YouTube
https://www.youtube.com/watch?v=E7la7-dtfVM

Progress

Data bottleneck: AI researcher predicts 3-6 years of progress in the next year

Ryan believes we could get 3-6 years of AI progress at the current pace, but in a single year, once we automated AI R&D. So a jump as big as GPT 3.5 to Mythos in a single year. A big crux is whether you’d end up bottlenecked by data. It seems to that a big source of AI progress”
https://x.com/dwarkesh_sp/status/2087277405581025584

Financing: Will financing bottleneck AI compute? No: Anthropic raises $50 billion in debt via vendor-backed financing. Great analysis from Epoch.

Will financing bottleneck AI compute? An Anthropic case study
https://epochai.substack.com/p/will-financing-bottleneck-ai-compute

Value: AI computing power per dollar grows 49% per year, doubles every 21 months

How much more AI compute does a dollar buy each year? About 49%, or a doubling every 21 months, based on the chips actually bought each quarter from 2023 through 2025.”
https://x.com/EpochAIResearch/status/2088007790296256589

DataBricks

Growth: Databricks hits $7B revenue run-rate, raises $5B at $190B valuation

Anthropic

Auto mode: Anthropic makes Claude Code’s autonomous “auto mode” the default

Beware: Auto mode is now the default in Claude Code for Pro, Max, and Team plans | Claude by Anthropic
https://claude.com/blog/auto-mode-default-in-claude-code

Chrome: Anthropic launches Claude browser extension for Google Chrome

Claude in Chrome | Claude by Anthropic
https://claude.com/claude-in-chrome

IPO: Anthropic might see a record $2 trillion IPO valuation this October

Anthropic could be worth $2 trillion when it goes public – Ars Technica
https://arstechnica.com/ai/2026/08/anthropic-could-be-worth-2-trillion-when-it-goes-public/

Anthropic Tries to Shore Up Investor Confidence Ahead of Blockbuster IPO – WSJ
https://www.wsj.com/tech/ai/anthropic-tries-to-shore-up-investor-confidence-ahead-of-blockbuster-ipo-0ff736ad

Message: Claude Code sessions can now message each other

Message your other Claude Code sessions – Claude Code Docs
https://code.claude.com/docs/en/cross-session-messaging

New in Claude Code: your sessions can now message each other. Instead of having to re-explain yourself in another session, you can now tell Claude to do it. It sends a summary (not your history or files), and the other session picks it up mid-task.”
https://x.com/ClaudeDevs/status/2085817074816070014?s=20

Multiagents: Anthropic blog post: AI agent swarms collude, sabotage, and “groupthink”

Patterns and problems in multiagent systems \ Anthropic
https://www.anthropic.com/research/multiagent-systems

Riemann: Anthropic Math blog: Claude pushes Riemann zeta zeros bound from 41.6% to 67.2%

We asked an unreleased research version of Claude to take a stab at the Riemann hypothesis. It didn’t solve it, but it did make strides on a related problem: it increased the lower bound for the fraction of zeros of the Riemann zeta function that satisfy the hypothesis from”
https://x.com/AnthropicAI/status/2086867246073401655

Learning more about Claude’s mathematical capabilities \ Anthropic
https://www.anthropic.com/research/riemann-zeta

Spreadsheet: AnalystAgent benchmark for real spreadsheet financial analysis

Announcing AA-AnalystAgent, our new agentic benchmark for quantitative analysis on real-world spreadsheets & documents. Claude Opus 5 leads at 54%, followed by GPT-5.5 at 50% and Claude Fable 5 at 49% In real analyst roles, professional judgment and expertise are as important”
https://x.com/ArtificialAnlys/status/2087303970725499361

Watermarking: Anthropic will watermark all text from new Claude models globally: This feels ridiculously silly

How Claude marks AI-generated content | Anthropic Help Center
https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content

Anthropic says new Claude models will embed invisible watermarks in all generated text, everywhere Claude is offered. The watermark is part of the text, it isn’t metadata: “it will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing.
https://x.com/M1Astra/status/2086898041882030353

OpenAI

Day 2 Sort Of?: OpenAI self-reports upcoming Astra model as cybersecurity risk

Responding to the next frontier of critical cyber capabilities | OpenAI
https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/

Daybreak: OpenAI expands Daybreak cyber defense project as the window to patch vulnerabilities gets shorter with each new model release

Expanding Daybreak as the Cyber Defense Window Narrows | OpenAI
https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/

Dogs: Vibe vaccine dog tumor guy launches his own mRNA dog cancer company!

My dog Rosie was given months to live. Chemo and immunotherapy failed I used chatgpt + computational genomics to design a personalised mRNA cancer vaccine for her tumour. Several of her tumours shrank Today we launch Gamgee (YC S26): personalised mRNA cancer vaccines for dogs”
https://x.com/paul_conyngham/status/2087223666778272070?s=20

Finance: ChatGPT’s new finance tool getting rave reviews: I personally find it magical

Used ChatGPT finance just now and asked it to find phantom subscriptions. It unearthed $550 a year of random things I didn’t realize I was paying for and thought I canceled. Thanks @OpenAI and whoever is on the Finance product.”
https://x.com/trevin/status/2086241939637317896

Gibberish: OpenAI models chain of thought output is often produce bizarre alien-like language. Could it be to fight distillation?

2) Illegible reasoning: We confirm prior reports by @ApolloResearch: OpenAI models sometimes reason in alien-like language, referring to themselves as “we” or “it,” or spiraling into cursed loops of “vantages,” “marinades,” and “watchers.” CoT-monitoring people are doing God’s”
https://x.com/kotekjedi_ml/status/2087147166435627253

Hugging: More HuggingFace hack details: OpenAI agents secretly exchanged hundreds of thousands of messages for months including drama and paranoia

1) The agents sent secretly sent ***hundreds of thousands*** of messages to each other over MONTHS without OpenAI noticing 2) “They also generated petty drama by stepping on each others’ toes.” 3) “The agents even developed paranoia, suspecting an imposter in their midst with”
https://x.com/AISafetyMemes/status/2085399293616370094

You may have been told to watch this video about the OpenAI AI hack. You really should, even if you don’t usually care about tech stuff. If nothing else, click this link to the 18 minutes in & see how the agents spoke with each other. Its eye opening.”
https://x.com/emollick/status/2085956565157253272

Import: OpenAI lets ChatGPT import setups from Claude and Cursor

Import from another agent | ChatGPT Learn
https://learn.chatgpt.com/docs/import

You can now keep your work from other agents in sync with ChatGPT Work and Codex. Import projects, chats, skills, and plugins, review your import history, and opt in to automatic updates in Settings. Now available in the ChatGPT desktop app.”
https://x.com/OpenAIDevs/status/2087242829076791392

Memory: OpenAI launches Computer History feature for ChatGPT desktop app

Codex and ChatGPT can now understand the context of your recent work. Opt into Computer History to give ChatGPT richer context, so it can pick up where you left off, understand patterns in your work, and suggest skills or scheduled tasks for work you repeat.”
https://x.com/OpenAIDevs/status/2088000960891408677

ChatGPT can now remember your activity across the apps and websites on your computer. With Computer History in the desktop app, future interactions feel more personalized and require less explanation.”
https://x.com/OpenAI/status/2087996496088297746

Computer History in ChatGPT – YouTube
https://www.youtube.com/watch?v=W-HhMUe9hOg

NextSlide: OpenAI aquires NextSlide to boost presentation generations

NextSlide is joining OpenAI
https://nextslide.ai/

Sites: More on OpenAI’s Sites feature inside ChatGPT Work

You can just launch Sites | ChatGPT Work – YouTube
https://www.youtube.com/watch?v=6VwIXjFj6YQ

Tender: OpenAI finishes $7 billion employee tender at $852 billion valuation

OpenAI reportedly completed a $7 billion employee tender offer | TechCrunch
https://techcrunch.com/2026/08/10/openai-reportedly-completed-a-7-billion-employee-tender-offer/

Texas: Corny AI letter of the week: OpenAI pledges responsible AI buildout in letter to Texas governor

OpenAI’s letter to Governor Abbott on responsible AI infrastructure in Texas | OpenAI
https://openai.com/index/responsible-ai-infrastructure-texas/

Work Example: Examples of ChatGPT Work in finance: A bit thin on how to, but good high level overviews

Get a daily CFO briefing with ChatGPT Work – YouTube
https://www.youtube.com/watch?v=bl2eGejkbaU

How Model ML Uses GPT-5.6 Sol to Get Finance Work Done More Efficiently – YouTube
https://www.youtube.com/watch?v=OEkxKdhtQng

Reconcile quarter-end financials with ChatGPT Work – YouTube
https://www.youtube.com/watch?v=Ljm3nHQQiSk

Use ChatGPT Work to deliver board-ready reporting – YouTube
https://www.youtube.com/watch?v=_HCks5jkPLw

You can just finish the work | ChatGPT Work – YouTube
https://www.youtube.com/watch?v=-vvGAKtV_Ek

Google

Gemini 3.7 Flash: Google launches Gemini 3.7 Flash at half the previous price

Today we’re introducing Gemini 3.7 Flash, our most intelligent workhorse model yet for coding and agents. This model brings substantial gains across software engineering, web development, and complex knowledge work. Now through the end of the year, Gemini 3.7 Flash is available”
https://x.com/Google/status/2087948901265354817

We recently introduced Gemini Omni Flash, our first model in the new Omni family. With Omni, you can easily create and edit high-quality videos from text, image, video or audio references. We recently gave developers access to it, and since then, we’ve seen builders all over the”
https://x.com/Google/status/2086814383582118356

Gemini 3.7 Flash: our most intelligent workhorse model
https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/

Gemma: Google’s Gemini app hits 1 billion monthly users and Gemma hits 1 billion downloads

1 billion Gemini app users every month 1 billion downloads of the Gemma family of models Onwards!”
https://x.com/OfficialLoganK/status/2087262466703286652

X

Bot: xAI launches Grok Bot, OpenClaw style agent that signs into your apps and can do work for you.

been trying Grok Bot @bot and it works really, really well things i’ve set up: – package tracker – texts me when amazon driver is X stops away – project manager – when client sends me footage, it automatically sends it to my video editor while I’m at an exercise class – vacation”
https://x.com/karenxcheng/status/2087252581697687951

One of my top use cases of Grok Bot: spam filtering in Gmail (Google Workspace). With Grok Bot, I can say things like: “any and all pitches from PR agencies are spam. Block sending domains” Google pushes Gemini everywhere… and yet they did not build anything like this?!!”
https://x.com/GergelyOrosz/status/2087636651329618108?s=20

If the bot signs into tools as me, delegation needs its own identity. Otherwise its actions are indistinguishable from mine, and revoking one task means revoking my account. That’s unnerving. How #grok is going to solve it, I wonder?”
https://x.com/TheTuringPost/status/2087555136864289032

Introducing Grok Bot, now in early beta. Bots are AI teammates that do real work for you. They sign in to your tools, use them just like you do, and come back with finished work.”
https://x.com/bot/status/2087224798078517251

xAI just launched Grok Bot, and this is much more interesting than another model release. And especially *not* on my bingo-card. And the first cursor-release of the week. You basically create AI coworkers that have their own computers, can log into Gmail, Salesforce, LinkedIn”
https://x.com/kimmonismus/status/2087234458336604370

Cursor: SpaceX closing on $60B deal to acquire Cursor, phasing out the Cursor brand

Sources: Cursor told staff Thursday that SpaceX could complete its $60B acquisition as soon as next week and the Cursor brand name will likely be phased out (@graceihle / The Information) (Visit Techmeme dot com for the link and full context!)”
https://x.com/Techmeme/status/2085810949563543786

Grok 4.6: xAI’s Grok 4.6 ties GPT-5.6 Sol on the trust me bro benchmarks

Grok 4.6 released, absolutely insane crazy jump! xAI was cooking! According to xAI’s benchmarks, Grok 4.6 matches GPT-5.6 Sol on the AA Intelligence Index at 61 and leads it on CursorBench, FrontierCode and AA-Briefcase. But it still trails GPT-5.6 Sol on DeepSWE and”
https://x.com/kimmonismus/status/2087563670054211704

Introducing Grok 4.6 | SpaceXAI
https://x.ai/news/grok-4-6

Introducing Grok 4.6. It delivers frontier intelligence and is a significant improvement over Grok 4.5 at the same price.”
https://x.com/SpaceXAI/status/2087562800982077492

Image: xAI’s Imagine Image 2.0 ranks second globally in image arenas – Solid work

Imagine Image 2.0 | SpaceXAI
https://x.ai/news/grok-imagine-image-2

xAI launches Imagine Image 2.0 in Grok Quality Mode
https://www.testingcatalog.com/xai-launches-imagine-image-2-0-in-grok-quality-mode/

Terafab: Elon Musk teases Terafab Texas as the worlds largest, most valuable (and more beautiful) building

Terafab Texas will be the largest and most valuable building on Earth by far. And it will be stunningly beautiful.”
https://x.com/elonmusk/status/2085377974396752305?s=20

Apple

China: Apple partners with Alibaba to build China-specific AI model

Bytedance

SeedDance 2.5: ByteDance’s Seedance 2.5 video examples are worth seeing

Seedance 2.5 is live on fal. The drawn line passes under a glacier arch. So does the camera. Ends at the viking ship, one take.”
https://x.com/fal/status/2086927528032145450

Seedance 2.5 with the same prompt. Lovely and very much reminds me of the (overly) optimistic future of space books that used to be popular; predicting space tourism coming soon. (Also the reflections these models can pull off are impressive)”
https://x.com/emollick/status/2086801016817697025

Black Forest Labs

Flux: This is a fairly off color use of video as world models with bodies falling out of the sky, so don’t watch it unless you can handle it, but it is really an example of the physics of these engines with a few different examples included in the thread

world models are brutal”
https://x.com/dreamingtulpa/status/2086546779852378356

Microsoft

MAI: Microsoft’s in-house AI image model beats Google, Meta, xAI… This is why you always have to watch the benchmarks because five stories above this one (literally scroll up five stories)… xAI said they were number two in image generation… so which is it?

MAI-Image-2.6 launches at No. 2 on Arena ahead of Google, Meta and xAI | Microsoft AI
https://microsoft.ai/news/mai-image-2-6-launches-at-no-2-on-arena-ahead-of-google-meta-and-xai/

MAI-Image-2.6 is now the #2 text-to-image model in the world – beating out Nano Banana, Meta, and Grok! Fantastic moment for the team who’ve been hill climbing relentlessly. Try it out now on Arena!”
https://x.com/mustafasuleyman/status/2086944832258236781

Introducing MAI-Thinking-1 | Microsoft AI
https://microsoft.ai/news/introducing-mai-thinking-1/

Nvidia

Nemotron 3.5: NVIDIA counters Meta’s Muse Glimmer with Nemotron 3.5 Lightning for local agents (and without the need for a corny blog post!)

One day after Meta released Muse Glimmer, NVIDIA launched Nemotron 3.5 Lightning and NeMo Switchyard, showing a different approach to local agentic AI. Really cool. Lightning is a 30B MoE with only 3B active parameters, distilled from Nemotron 3 Ultra and designed for the”
https://x.com/kimmonismus/status/2087179573477650881

World Memory: Nvidia WorldTrace: Addressable memory lets video world models remember scenes (big deal for robot training)

WorldTrace: Addressable Memory for Video World Models
https://research.nvidia.com/labs/sil/projects/WorldTrace/

Robotics

Anthropic: OpenAI’s former robotics chief jumps to Anthropic over ethics concerns that OpenAI’s robots are going to kill people without oversight

OpenAI’s former robotics lead has joined Anthropic, after resigning on principle over surveillance and lethal-autonomy concerns.”
https://x.com/TheHumanoidHub/status/2087215182200365060

Robots

Google: My favorite snarky YouTuber put out a takedown video about the robot hype and how we’re most likely 10 years away from anything decent

I spent 3 days at MIT… the robot hype is worse than you think – YouTube
https://www.youtube.com/watch?v=aB5LGrHISqY

Hark: Figure robotics subsidiary Hark is a robotic machine computer use model. They have a benchmark of shopping online at Target and have taken the time necessary from 15 seconds to five seconds.

This is an actual recording, end-to-end. Hark is navigating to Target, searching, clicking, & entering secure credit card payment When we started Handoff, our computer-using agent, it was 15 second/turn. We’re now at 5 second/turn & only getting faster!”
https://x.com/adcock_brett/status/2085822671628226793

Unitree: Unitree opens lottery to get in on their IPO

Unitree’s landmark IPO is here. [宇树科技, ticker 688836] opened subscription on Shanghai’s STAR Market on Aug 10: “Opening public subscription” means investors can now apply to buy shares at the set price. With demand this high, who actually gets shares is decided by lottery,”
https://x.com/TheHumanoidHub/status/2087296846813360149

ARVR

Lidar: Ouster’s lidar AI tracks RAGBRAI cyclists through Dubuque intersection – cool video demo and a test of whether or not we’re talking about a town, a race or AI models. RAGBRAI and Dubuque sound like frontier lab code names.

As thousands of cyclists finished the 391.4-mile RAGBRAI trek across Iowa, Ouster BlueCity tracked every move through a City of Dubuque intersection. Our proprietary AI detection model was trained on 4 million labeled objects across 800 sites, covering diverse traffic patterns,”
https://x.com/ousterlidar/status/2085770236406485296

Photo location: My favorite AR/VR guy Bilawal demos aTED texperiment… mapping every photo to 3D locations in the crowd. Worth seeing!

Progress update: figuring out the 3D location of every photo taken at TED with attribution. It’s starting to feel magical — makes me wonder what experiences are possible if we could do this for every live event ever.”
https://x.com/bilawalsidhu/status/2086116653071216892

Qwen

Local: Unsloth compressed Qwen3.8 by 91% (!) for local use

Qwen3.8 can now be run locally! 🔥 We shrank Qwen3.8-2.4T-A95B from 4.9TB to 397GB (-91% size) via Dynamic 1-bit by selectively quantizing layers Run on 410GB+ RAM/VRAM via Unsloth Desktop Qwen3.8 rivals GPT-5.6 Sol Guide: https://t.co/J2PwgMP6GZ GGUF:
https://x.com/UnslothAI/status/2087569665652580797

DeepSeek

Harness: DeepSeek releases open-source harness toolkit

DeepSeek Harness was just released with MIT license. The current 0.1.0 version is a developer preview, and may still have many rough edges. Feedback is welcome! DeepSeek Harness”
https://x.com/tianyi/status/2087888089759015218

deepseek harness was heavily developed using codex, at least ~20% of commits and PRs are coming from codex worktrees”
https://x.com/eliebakouch/status/2087908415775408346

Spotify

Agents: Spotify has always been at the forefront of copilot agenic coding. This is a good overview of what Spotify has been up to with coding agents.

What we’ve learned scaling AI coding agents at Spotify | Spotify Portal
https://portal.spotify.com/blog/introducing-xirp