About This Week’s Covers

This week’s cover celebrates the life of David Hockney, the British painter and photographer who died on June 11, 2026. I chose a 1977 painting called My Parents. I swapped out the book for an iPhone and transformed his mom into the Figure 03 robot.

Part of my newsletter’s goal is to draw awareness to the humanities. I’ll be talking about AI, but along the way, I will learn about artists, writers, and other creative people.

I’ve included a few of my favorite category covers below, created by Claude and Gemini without my guidance.

This Week’s Humanities Selections

This week’s Humanities readings are quotes by David Hockney that I felt were relevant to AI, life, and expression.

“The moment you cheat for the sake of beauty,you know you’re an artist.”
― David Hockney

“Most art will disappear. The past is edited so it always looks clearer to us. Today always looks a bit of a jumble. We’ll put up with rubbish from now, but not with rubbish from the past.”
― David Hockney

“The world is very, very beautiful if you look at it. But most people don’t look very closely. They scan the ground ahead of them so they can walk, but they don’t observe things very attentively, intensely. I observe, & I’ve always known that.”
― David Hockney

I’m also adding a song of the week: something uniquely human, chaotic, or organic. When possible, I’m tying it into the date range.

This week’s song is “Are You Glad to Be in America” by James Blood Ulmer, who died this week on June 3, 2026.

Ulmer was known for using simple, unconventional guitar tunings, sometimes just two notes across all six strings, and even single-pitch tunings! He would change the gauges of the strings to give them different resonance even thought they were all the same note.

Ulmur played with Art Blakey’s Jazz Messengers and Ornette Coleman, another eccentric musician who I featured in a previous newsletter. I had never heard of Ulmur until this week, and I’ve enjoyed listening to his music.

This Week By The Numbers

Total Organized Headlines: 584

This Week’s AI News Overview In Plain Terms (written by me, so it’s rambling and not AI, you’re welcome)

This week I hand-organized 584 links into about 60 categories. A whopping 123 links went into the executive summaries. I’m 12 weeks behind because I enjoyed the summer with my family.

Touching grass is worth it.
The weekend of June 12th was the air show in Ocean City. I used to go out on the beach and take pictures, but this year I stood in our driveway. We’re lucky that our house is on the 50-yard-line approach as the planes head toward the beach, and I can get some pretty neat pictures.

I continued training for my attempt to climb to the top of Grand Traverse Peak outside of Vail, Colorado, with my daughter in August. I did a StairMaster workout and climbed 275 flights with a 20-pound backpack. That’s 2,958 feet of elevation gain. Grand Traverse is a 13,063-foot mountain, and the hike starts at 8,120. That’s just shy of 5,000 feet in gain for the hike. Living at sea level with no hills, I do the StairMaster and soft-sand runs as proxies.

Climbing 275 stories with no break is insane. Only 2000 more vertical feet to go…

Let’s get into this week’s AI news
I have organized this week’s links in groups by company. I jump around, but the first set will be what I consider the top news, and then we’ll move through.

There are about 48 executive summaries this week. Some of them are huge in the news, but I didn’t make them the top AI story. For example, the SpaceX IPO is news, but I think it’s more important to talk about the big headlines in AI that are under the radar.

I try to make this newsletter a note to myself, so when I look back, these are all the things that I noticed from the week. I hope it’s helpful, and there’s a ton of links and supporting documents and imagery along the way.

The top story of the week was that Anthropic has finally released its much-awaited Fable model. Fable is a nerfed version of their premier model, Mythos. Mythos is an extremely powerful model that was able to find security vulnerabilities across almost every system imaginable (finance, government, corporations, software). Anthropic did not want to release Mythos because it presented security risks, so they spent time creating protections. A lot of folks called it marketing BS… I did not.

The Mythos model is only available through a private release to a limited number of partners… called Project Glasswing. These folks are kicking the tires on Mythos and using it to strengthen their computer systems so that, when Anthropic does release the next version, all of the systems have been patched before the public gets access to this powerful hacking tool. That’s relaxing.

There is a lot of debate and skepticism around almost every AI announcement of capabilities. But I listen to the same vetted voices week over week, and the more sober people I follow have expressed that Mythos is incredibly powerful, and Fable is as well. Fable is just not able to do as many black-hat activities like hacking. Anthropic released a little video that has an overview of this. They also released some benchmarks, which I’ll share below in the full list of headlines.

Mythos was able to speed up its training about 69 times as fast as before. People have said that a human expert needs about four to eight hours to get a four-times increase in speed. AI is going to start improving itself.

With Fable, the chain of thought is starting to become gibberish. It thinks for a long time, and it starts to build a shorthand for what it’s doing without realizing that people can’t read it. This is going to be a challenge, as chain of thought, which is thinking out loud, step by step, is important to keep track of a model’s alignment. In the past, people thought perhaps the AI would lie or that chain of thought wasn’t representative of what it was doing, but it was better than nothing. Gibberish and code words are not what people want to start seeing emerge from these technologies.

Another big deal with Fable is that it can orchestrate and create agents.

Anthropic has a platform called Claude Managed Agents, and that’s an environment you can use to create agents and delegate work among them. It’s outside the realm of laypeople, myself included, but Fable is able to connect to it to delegate work to agents without the user needing to know how to set them up.

The reason this is important is not just because it’s delegating work to something else. But the second reason is that the agents Fable delegates to can use cheaper models. The higher the model you use, the more powerful the brain it has, the more tokens it uses, and the more expensive it gets.

So, just like in real life, let’s say you had a golf cart and a Ferrari. If you’re going to check the mail, you might take the golf cart. If you want to go for a fun ride, really fast, you’d use the Ferrari. But you wouldn’t want to be wasting gas or using a tool that was overqualified.

It saves a ton of money if you can delegate tasks from Fable to managed agents that use smaller models. We’ll see a lot more about it in the next week or two.

There are a lot of examples of the power of Fable in the links below. You’ll see examples of people saying that they asked for an impossible task, and then Fable will work for nine hours. It can also use tools. One guy had it build a working mechanical engine using CAD.

At this point, everyone is testing and learning what it can do. A lot of the feedback is that we have to think of things that are beyond our assumptions of what it can do and then see when it succeeds. It’s like drilling for oil.

Claude is now writing 80% of its code. Self-improving technology means that the pace of change is going to increase…. Ray Kurzweil may finally live to see his moment of the singularity.

Anthropic put out a compelling article that talks about the recursive self-improvement that may occur in the near future when AI starts building itself.

Must read: https://www.anthropic.com/institute/recursive-self-improvement

Developers have started to use the term “loops.” Instead of writing prompts, they create loops. You’re going to hear about this more and more.

Loops are a series of broad commands, like build something, test something, confirm it works, go back, fix mistakes, start the next process. It’s nested delegation, and it’s more complicated than basic prompting. The term has started to appear in a lot of developer commentary, and we’ll keep track of it over the next few weeks.

Dario Amodei let another lofty essay loose called “Policy on the AI Exponential.” https://darioamodei.com/post/policy-on-the-ai-exponential

TLDR: Technology is moving faster than people can understand it, and the people in charge of making the policies are the least informed.

Dario suggests AI model regulation around agencies like the FAA. He also says that the government should be able to stop the deployment of any model that it deemed too risky.

It’s wild to me that Dario would suggest that the government should have the power to block or deter deployment, given that Hegseth and Trump declared Dario a supply chain risk.

Dario talks about using AI to speed up science and medicine. He then talks a lot about protecting civil liberties. I’m not sure if AI is just the cherry on top of an already rough state of affairs when it comes to surveillance, but clearly AI takes surveillance to a next level.

Dario argues that democratic countries need to beat China and he implies that we are in an arms race.

Just a few weeks ago, Anthropic and the U.S. government were going about as aggressively head-to-head as humanly possible, to the point that the government declared Anthropic a supply chain risk, which is essentially an enemy of the state. It’s decimating for business, because not only can the company no longer work with the U.S. government, but any company who works with the U.S. government can’t engage Anthropic in any capacity.

All it took was Mythos to show up. Anthropic was able to crack security for almost every agency, bank, and platform across the country. Anthropic’s security hacking ability is too strong for the U.S. to ignore.

So now, even while they’re in the middle of lawsuits with the Pentagon, six Anthropic engineers are working inside the NSA to deploy Mythos for offensive operations.

Anthropic made a few major mistakes when it nerfed Mythos to build Fable. Mainly they tried to avoid biochemical research, but rather than create hard blockers on queries and denials, they redirected Fable queries down a step into the less capable model Opus without disclosure.

Thousands of scientists and researchers in the United States wanted to try Fable to see how it would do for their studying, and without telling them, Fable started deprecating them down to Opus.

Anthropic reversed this within a few days, but it all comes back to the question of transparency. If this is the stuff we know, I can only imagine the things we don’t know.

Let’s switch over to Apple’s announcement of a small locally hosted model.

Many people consider AI a monolithic term, and they don’t think about choosing different models for a different task.

Developers need to think about this more than anyone else, because rather than using the chatbot at a fixed fee per month, they’re using a pay-as-you-go, token-based system, and the model they choose is correlated to how much money they spend.

If you have a simple task, like extracting a name out of an email, you can use a very small model. But if you need to track every player in a three hours sporting event, you need an expensive multimodal model.

There’s also the question of whether the model lives in the cloud or locally on your device. If a model can be hosted locally on your laptop or your phone, you don’t need an internet connection to use it. The model has to be extremely small, and the smaller the model, the less powerful it is, because it doesn’t have access to all of the cloud compute and the data centers.

If you’re Apple or Microsoft, the more you can fit on a device, the faster it is, the more private it is, and the cheaper it is. If certain parts of your interaction take place on that device without having to hit the cloud, you could talk to your phone or your computer and get answers immediately, as long as your question wasn’t too complicated.

This week, Apple’s new 20-billion-parameter on-device model is a breakthrough in putting more onto a local device than ever before.
https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models

Another big update from Apple this week was the addition of 3D Gaussian splatting into 300 cities on Apple Maps. This may not sound like a lot to laypeople, but they can extrapolate detail into maps by stitching together the correlations of vectors from fixed imagery. So you’re taking several points from several images and then figuring out what the points between them would be, and that creates high resolution maps that are mind blowing to browse.

In Google news, Google announced that AI Studio is now generating over a million apps per week and has had over 18 million apps created since February. I don’t even know what Google AI Studio is, which is ‘shame on me’. It goes to show how much there is to follow. Google also released a ton of updates to Notebook LM, which I have stopped using personally. Google’s really falling apart (for me). I don’t use Nano Banana anymore either.

Google also announced they are going to rent computing power from SpaceX for $920 million a month. SpaceX is becoming the AWS of chip rentals for so many big companies, including Anthropic and now Google.

Google’s Gemma is a strong multimodal model, even though its on-device version is only 4 billion parameters (compare it to Apple’s 20 billion this week). This week Google launched Diffusion Gemma, which is an interesting angle on text models. Most language models do next-token prediction. This diffuses huge chunks of text almost like it would an image, as a whole block. So instead of breaking down letters and words and grammar, it just builds huge blocks of text at once.

If nothing else, for things that aren’t too serious, I think text diffusion makes a lot of sense. Greetings, salutations, simple things might as well be diffused. They’re boilerplate anyway.

I’ve been following humanoid robot development for about three years, and if you had asked me three years ago, I would have said that the X1 Neo would have been the frontrunner, or maybe Boston Dynamics, and then Unitree. However, for a while, Figure has been dominating.

Tesla Optimus is behind the competition, and this week BYD has announced they are building humanoid robots, using auto dealers for distribution. BYD is coming after Tesla in a big way, and now with robots, they’re doubling down.

China has announced plans for a $295 billion AI infrastructure build. China is racing forward quickly, whether it’s trying to find ways to not rely on our chips, building its own data centers, or flooding the market with open source. China is full out, pedal to the metal.

This week China has approved the world’s first brain-computer interface for commercial use.

Chinese open source model Kimi launched a locally run agent that can leverage 300 agents working at the same time on a local machine. It includes browser use too.

Switching to OpenAI, ChatGPT added an ability to read and create charts and process data within their apps. This is a bad sign for business intelligence jobs. We can take giant piles of raw data and give it to a language model, and GPT will do as good of a job as a business analyst, if not better.

OpenAI, much like Anthropic, created an economic research division that looks at the impact on businesses and jobs. That’s new this week as well.

OpenAI’s API can now include image results when searching the web.

In existential news, OpenAI has introduced Lockdown Mode that will hopefully stop prompt injections. The more AI can use our computer and browse the web for us, the more susceptible it is malware. Let’s say you delegate a task to ChatGPT. It starts opening up a bunch of websites and searching the web on your behalf, and then along the way it grabs some malware, and there’s a prompt injection to take over your machine. Lockdown Mode would shut the whole thing down if it senses anything weird.

As of this week, ChatGPT can maintain its context and memory across chats. Some people are anti-memory, but I want it to have the full context of all my conversations. The more you talk to a chatbot, the less you want to repeat yourself or lose the context of a previous project.

This week Visa partnered with OpenAI to enable AI agents to shop.

An interesting product this week is ChatGPT’s launch of Sites. For a long time now, Codex and Claude Code could build websites that are saved locally on your hard drive as HTML files, but getting them into the cloud required some sort of server provider. Now ChatGPT will host these sites in the cloud. So you can build and publish little web apps and vibe code and then share the results with friends. I’m not sure how extensible it is, but it’s a bridge for laypeople.

Not to be left out, Sam Altman has penned his own crazy self-important essay this week! https://openai.com/index/built-to-benefit-everyone-our-plan/

In this case, Sam’s comparing AI to electricity. The essay has three main points. OpenAI wants to build automated AI researchers so that humans can focus on other things. Sam wants to try to boost economic gains, and he wants to give everybody on the planet a smarter-than-human personal assistant. It’s important to read these articles for a lot of reasons.

One of the largest stories of the week generally is the SpaceX IPO. There are all sorts of red flags and callouts and constraints. The links below are worth reading.

Speaking of IPOs, Anthropic has filed confidential drafts to proceed with their S-1 registration.

OpenAI also filed.

The fact that OpenAI and Anthropic filed in the same week is just kind of bonkers.

OpenAI is also looking to build a 10-gigawatt data center in Ohio, backed by Nvidia. The circular investing just won’t stop. You just wonder if all this money and risk can successfully skim across the surface like a rock and make it to the shore before it sinks.

A report asserts that OpenAI and Anthropic may have to spend over $1,000 for every $100 they earn.

Andrej Karpathy is one of my favorite voices in artificial intelligence, primarily because he’s so generous with his time and he’s a very even-keeled guy.

On the Dwarkesh podcast, Andrej said he thinks AGI’s impact on the economy will be indistinguishable from the current rate of growth. He doesn’t think AI’s impact going to show up in gross domestic product.

This is a counterpoint to Dwarkesh, who says that AGI is going to cause a massive jump in productivity. I tend to go with Andrej with almost anything he says (based on his record). We’ll have to follow along to see if he keeps this point of view.

Marc Andreessen, who I disagree with on almost everything, has come out with thoughts on the death of the web browser and why he doesn’t think AI is a bubble. It’s easy to dismiss him because he’s a venture capitalist. I would recommend skimming the transcript, if nothing else.

If you want a more down-to-earth version (or middle ground) of Andrej Karpathy and Marc Andreessen’s thoughts, The New York Times had a roundtable debate on AI’s impact on workers. This is way more in layman’s terms. It will catch you up quickly if you just want to read one article.
https://www.nytimes.com/2026/06/09/magazine/ai-jobs-workforce-labor.html

In science news, the Mayo Clinic is now able to catch pancreatic cancer three years before it would usually be found on routine CAT scans.

In ethical news, Martin Scorsese is in the hot seat, as he came out with a partnership with Black Forest Labs (for AI video).

Lionsgate announced that it’s going to take an equity stake in Runway, another AI video platform.

Speaking of AI in films, the state of New York now requires that any ad that has an AI-generated performer include a disclosure that identifies the ad as AI.

And last for this week, with the World Cup coming, FIFA has announced strong AI tools to help with officiating and analysis. So whether you’re a referee or a fan or a color commentator, AI can now track players, follow along, and help with decisions, especially things like offsides or instant replays.

This Week’s Top Stories with All The Links

Anthropic

Fable: Anthropic’s SuperAI Claude Fable 5 (aka Nerfed Mythos) is here

Claude Fable 5 and Claude Mythos 5 \ Anthropic
https://www.anthropic.com/news/claude-fable-5-mythos-5

Claude Fable 5 is our first generally available Mythos-class model. It ships with new safety classifiers that may flag certain prompts in dual-use domains like cyber and bio. We’ve added fallbacks: a refused request retries on Claude Opus 4.8 instead of dead-ending.
https://x.com/ClaudeDevs/status/2064428347678220691

Introducing Claude Fable 5 – YouTube
https://www.youtube.com/watch?v=Y9Wz2PV404E

my response to fable was to be quiet, still and humbled. i felt a panic that i didn’t have purpose anymore. it strictly dominates me as an engineer. i had a month of that before realising that more than caring about “how i achieved my goals”, i just cared about “my goals” and a
https://x.com/akbirkhan/status/2064418425552928812

My take 24 hours after Fable 5: Your organization will likely not scale with the exponential curve of AI. I’l just come out to say: This should be a wakeup call for engineering teams. Set up your cloud software factories. Now. Models can now fix impossible bugs, UI-test the
https://x.com/walden_yan/status/2064755974548902006

Mythos 5 speeds up its own training by over 69 times a human expert needs 4–8h eq. for a mere 4x speedup
https://x.com/scaling01/status/2064392809293939119

One interesting pattern with Fable 5 is that it will often say things that are gibberish when I use it for coding. Things like “”The morning’s slim-scan fix cured the scan hang””, “”this is a latent-drift API-shape wrinkle””, etc. When I ask why it does this, Fable explains that it
https://x.com/tamaybes/status/2065147305494450248

One thing I mentioned only in passing in my Fable post is that, for long running tasks, Fable starts to develop its own dialect as its many agents and tasks reinforce themselves and make Claudish language ever more Claudish. You need to ask it to report out in plain English.
https://x.com/emollick/status/2064542441848422611

With multi-agent orchestration in Claude Managed Agents, you can use Fable to delegate work to dedicated agents using smaller models.
https://x.com/ClaudeDevs/status/2064394928948703406

Anthropic has released Claude Fable 5, the first publicly available Mythos-class model that ranks #1 in our agentic real-world knowledge work benchmark GDPval-AA Claude Fable 5 shares the same underlying model as Claude Mythos 5, with added security guardrails for potentially
https://x.com/ArtificialAnlys/status/2064414308289937869

Anthropic’s new Claude Fable 5 is #1 on Terminal-Bench 2.1 at 88.0%, beating GPT 5.5 by 4.6%. Under the hood it’s their Mythos model with safeguards that route dangerous requests to Opus 4.8, which Anthropic claims triggers in under 5% of sessions. Available in Cline now!
https://x.com/cline/status/2064427461212045546

Claude Fable 5 / Mythos 5 wins everywhere. I thought Fable 5 was just a nerfed Mythos Preview, but it’s literally better. SWE-Bench Pro: Fable 5: 80.3%, GPT-5.5: 58.6%. And the price is only 2x Opus 4.8: $10/input MTok, $50/output MTok. I don’t think GPT 5.6 can beat this…
https://x.com/Yuchenj_UW/status/2064396097075003739

Claude Fable 5 beats Pokémon FireRed only using vision – YouTube
https://www.youtube.com/watch?v=Ty_50J84fMY

claude fable 5 has solved CAD I asked it to make a model of a V8 engine It came back to me with a fully working model in under 10 minutes
https://x.com/aaronli/status/2064876123109089742

Exciting news: Claude Fable 5 ranks #1 on the new Agent Arena leaderboard! Fable 5 leads by the widest margin ever over Opus-4.8 and GPT-5.5 on two key signals: confirmed task success rate and praise vs. complaint, despite weaker steerability. If Fable can do something, it will
https://x.com/arena/status/2064807170714358193

Fable 5 is the biggest step up I’ve felt in our models since Opus 4.5 back in November. After 4.5 came out I uninstalled my IDE when I realized that I’d been doing 100% of my coding in a terminal for a few weeks. With Fable, it’s felt like Claude has stepped up from being a
https://x.com/bcherny/status/2064431111154053187

Fable 5 lies 96% of the time. We were surprised by it’s skill… 🧵
https://x.com/kradleai/status/2064907897373642912

Fable 5 scores 81.9% on SimpleBench the highest score almost reaching human baseline.
https://x.com/JasonBotterill/status/2064699951578505446

Fable: “”create a visually interesting shader that can run in twigl-dot-app make it like an infinite city of neo-gothic towers partially drowned in a stormy ocean with large waves.”” “”Make it better”” All of this is procedurally generated.
https://x.com/emollick/status/2064424775527624736

Fable: “”write me a rhyming poem with six four line stanzas, each stanza removes another vowel. the first has no u, the second no u or i, etc.””
https://x.com/emollick/status/2064769268013584767

I think we’ve reached the point where normal people can’t really determine whether new models are better than previous ones. Like Fable doesn’t seem that much better to me, but every 150 IQ person I know is like “wow the singularity came sooner than I thought”.
https://x.com/citrini/status/2064480613852201336

I’ve had access to Fable for a bit. A genuine jump in capability, I could feed it a 15 page design document for a project and it would work for 9+ hours and deliver terrific results. But working with it is weird & weirder is coming Lots of examples:
https://x.com/emollick/status/2064395281903346013

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any model we’ve ever made generally available.
https://x.com/claudeai/status/2064394146916229443

Just used claude fable (aka mythos) to create this city block simulator complete with multi-agent traffic, live detection boxes + tracks, and day to night cycle. And it just one shotted it. This is gonna be fun — the gap between idea and execution just keeps collapsing.
https://x.com/bilawalsidhu/status/2064524211914223867

More realistic example of a one shotted game. Asked Fable 5 to recreate a game in the style of The Elder Scrolls 5 Morrowind. It one shotted quests, currencys and fighting, journal and minimap. And it worked.
https://x.com/kimmonismus/status/2064744343349399634

New #1 on PostTrainBench: Opus 4.8 (max reasoning) hits 37.23% — up from 28.56% for 4.7, the largest single improvement we’ve seen. Fable 5 runs underway now that AI research behavior is no longer silently degraded. PostTrainBench asks how well frontier AI can train weaker
https://x.com/thoughtfullab/status/2065096885514227876

This is a super exciting release – Claude Fable 5 is the same underlying model as Mythos but with added safeguards. The benchmarks are great and it’s SOTA on everything by a margin but I’ll add that *qualitatively* also, this is a major-version-bump-deserving step change forward
https://x.com/karpathy/status/2064409694761054332

Want to be very upfront about the subscription rollout (Pro, Max, Team, seat-based Enterprise). We’re giving users Fable 5 within subscription limits for 2 weeks, and then we’re taking it away. Here’s exactly what’s happening and why:
https://x.com/TheAmolAvasare/status/2064393574431764928

Impact: Claude writes 80% of its own code now (more in a bit below)

As of May 2026, more than 80% of the code we merge into Anthropic’s codebase was authored by Claude.”” Matches independent measures. There really is no sign this is slowing down (which doesn’t mean there aren’t organizational challenges to absorbing this much productivity gain)
https://x.com/emollick/status/2062580212768936344

Head of Claude Code said the line that makes prompting feel outdated “”I don’t prompt Claude anymore. I write loops, the loops do the work.”” Most devs are still trying to write the perfect prompt Boris Cherny already moved to the next layer Not prompt Not answer Not copy paste
https://x.com/0xwhrrari/status/2064804504608887040

Essay: Dario from Anthropic write another boorish essay that I mostly agree with (despite itself) – TLDR “We can’t keep up with this stuff. We’re toast”

Dario Amodei — Policy on the AI Exponential
https://darioamodei.com/post/policy-on-the-ai-exponential

Policy on the AI Exponential \ Anthropic
https://www.anthropic.com/policy-on-the-ai-exponential

AI is advancing at a pace our policymaking institutions were never built for—and the gap between the two is becoming the central challenge of the technology. In his latest essay, our CEO Dario Amodei lays out how to close it. We’re launching three new initiatives to support the
https://x.com/AnthropicAI/status/2064783418844762489

Today I’m publishing a new essay, Policy on the AI Exponential. AI is progressing extremely fast—much faster than the policy process was built to handle. The essay lays out where I think the technology is now, and the action needed to close the gap:
https://x.com/DarioAmodei/status/2064781775247950326

Charts: Anthropic chart maps when to use AI agents versus workflows (and it’s already outdated)

This chart from Anthropic is useful, since Agent Teams and Workflows are both very new and very powerful (and token hungry). On the other hand, maybe it doesn’t matter as a lot of the decisions about which approach to use is from the AI itself & it often uses them in combination
https://x.com/emollick/status/2063073955968123062

RSI: Anthropic: When AI Builds Itself (this is an important one to read)

When AI builds itself \ Anthropic
https://www.anthropic.com/institute/recursive-self-improvement

I think it is really worth reading this piece on RSI at Anthropic. There is a bit of navel-gazing, some marketing, and a lot of very sincere beliefs about what Anthropic thinks is likely in the near future of AI that you probably want to be aware of.
https://x.com/emollick/status/2062582362194460698

Moat: Anthropic is CRUSHING everyone on coding right now

Anthropic has a coding MOAT
https://x.com/scaling01/status/2064399642603802676

The capabilities of Claude Code and Codex have expanded a lot in recent months, they added many ways to approach work (subagents, skills, goal, workflows, plugins, etc). Given the AI labs can use their own AI to help documentation, a surprising amount is effectively undocumented
https://x.com/emollick/status/2062510975513747606

Mythos: Anthropic partners with the NSA because AI’s security hacking is too good (TLDR: stops fighting with Hegseth because sh*t got real)

Anthropic Embeds Engineers in the NSA to Deploy Mythos
https://www.implicator.ai/anthropic-embeds-engineers-in-the-nsa-to-deploy-mythos-for-offensive-cyber/

Claude Mythos 5 speeds up kernels by 430x – a 300x speedup equals 40 human expert hours of work
https://x.com/scaling01/status/2064392386520780945

Mythos 5 accelerates aspects of drug design by 10x
https://x.com/scaling01/status/2064394250142265367

Claude Mythos 5 scores 161.29 on Anthropic’s ECI
https://x.com/scaling01/status/2064392088003756431

Claude Mythos is next level. h/t @Lentils80 Look at this MacOS output. One shotted.
https://x.com/kimmonismus/status/2062843119864021404

Claude mythos will be on a completely different level. These outputs are insane
https://x.com/kimmonismus/status/2062805570982203820

Policy: Anthropic was open about being sketchy with researchers, and gets bodyslammed, reverses policy (did they?)

Anthropic Walks Back Policy That Could Have ‘Sabotaged’ AI Researchers Using Claude | WIRED
https://www.wired.com/story/anthropic-responds-to-backlash-on-claudes-secret-sabotage-on-ai-research/

Anthropic’s new Fable 5 safeguards are fascinating. When the model is used for frontier LLM development, it apparently does not simply refuse or warn the user. Instead, it quietly limits its own effectiveness through techniques like prompt modification, steering vectors, and
https://x.com/kimmonismus/status/2064417460715962479

If Claude Fable stops helping you, you’ll never know — Jonathon Ready
https://jonready.com/blog/posts/claude-fable5-is-allowed-to-sabotage-your-app-if-youre-a-competitor.html

When Fable 5 is used for frontier LLM development, it does not notify the user and instead limits the model’s capabilities through methods such as prompt modification, steering vectors, and PEFT. Anthropic estimated that this would affect approximately 0.03% of traffic.
https://x.com/Hangsiin/status/2064397550434816088

Cybersecurity and biosecurity requests may auto-reroute to Opus 4.8 (shown in the UI, billed at Opus prices). Docs + prompting guide:
https://x.com/ClaudeDevs/status/2064394931033248226

I am curious if the new Fable safegaurds would trigger accidentally on meaningfully important work as a false positive that has a real life consequence. Just realizing this is eroding trust and please change this to just down right blocking which is a fine position to take.
https://x.com/_arohan_/status/2064644778147643401

I believe what Anthropic is doing, gating the ability to do certain harmless things like LLM research, and with incredibly sensitive filters that even medical questions are often blocked, is *deeply* wrong. They got open research, the Transformer, GPT2, …
https://x.com/antirez/status/2064766431531532588

I don’t really want to have to go to bat against Anthropic, but they’ve just been unnecessarily antagonistic to all of China, then not so subtly to open weight models, and now more broadly open AI research. What’s next on the list?
https://x.com/natolambert/status/2064412173527556298

I got a good nights sleep and I’m still just as angry about Anthropic’s choices. I enjoy working in AI so much and to have my access to the cutting edge models for my work rugpulled in an under the table fashion is appalling. I expected to be restricted eventually, but not
https://x.com/natolambert/status/2064699044145095104

I’ve been testing something after @OliviaHelenS noticed you can’t even say “”Hi”” to Fable if you’re a biologist. I checked, and several of us are able to interact with Fable in Incognito Mode, but not in normal mode. This didn’t happen to our non-biologist friends.
https://x.com/cremieuxrecueil/status/2064449457869984035

In good faith and with no judgment (mistakes happen), I truly hope that Anthropic will hear the feedback and change course on this. Anthropic is a company that has been raising awareness about AI manipulation which is a very important topic! You don’t want to go down as the
https://x.com/ClementDelangue/status/2064673792303955985

It’s state of the art on nearly every benchmark we tested and the lead grows the longer the task. Made safe for general release: cyber & bio requests fall back transparently to Opus 4.8 and 95%+ of sessions never see one. $10/$50 on the API, in paid Claude plans today.
https://x.com/mikeyk/status/2064392996288901392

Makes me wonder how long this has already been going on without users being notified. I had been feeling like codex was running circles around Claude code for months now. Now I wonder if Claude code was just self nerfed. Regardless of the intentions behind this, this is a bad
https://x.com/code_star/status/2064464447662707180

My last observation re: Anthropic’s secret sabotage safety policy, is that it undermines actually good safety policy. How? 1. First, it is very plausible to describe this as anti-competitive behavior (even if you are maximally sympathetic to Anthropic here you must admit this),
https://x.com/deanwball/status/2064665679307985244

Re the Fable ML sandbagging, the model’s AI research capabilities were probably at least partly trained on Anthropic employees diffing atop proprietary algos and infra. So the IP leak is somewhat like a researcher who knows Anthropic’s stack getting poached to another lab.
https://x.com/dwarkesh_sp/status/2064826554442719502

SITUATION UPDATE: Anthropic is reversing its Fable 5 policy of covertly degrading performance for competing AI researchers, per Wired.
https://x.com/MTSlive/status/2064922000020398331

That was quick: Anthropic reversed a controversial policy that would have secretly degraded Claude Fable 5 for users doing frontier AI research after backlash from researchers who saw it as covert sabotage of competing AI development.
https://x.com/kimmonismus/status/2065003618710008084

The right term for this is supply chain attack. Depending on your use case, the Fable model will be a malware.
https://x.com/deliprao/status/2064485687374569897

Things I really dislike about Fable: 1. Anthropic collects my prompt history, stores it, and does whatever they want with it for 30 days. No opt-out 2. They can nerf their most expensive model without telling me, billing me the same amount, wasting my time. Whenever they want
https://x.com/GergelyOrosz/status/2064618497150210391

Very pleased to hear Anthropic have walked back this policy
https://x.com/simonw/status/2064918665859080392

why not just refuse the prompt? why so sneaky?? @AnthropicAI
https://x.com/DBahdanau/status/2064692204287799728

Apple

Local: Apple delivers 20B-parameter on-device model (big deal)

It’s very cool that Apple shipped a 20B parameter on-device. You can’t put 20B parameters in RAM at any reasonable precision. To make it work they are using pretty exotic architecture by today’s standards. A small model predicts from the query (or prompt) which experts to load
https://x.com/awnihannun/status/2064202168618422396

Maps: Apple launches 3D Gaussian splat maps for 300 cities (crazy high res maps)

Apple Beat Google to It! Hyper-Realistic Maps at WWDC 2026 — Gaussian Splats & the ongoing 3D maps land grab to own the substrate that connects the physical and digital world. Hopping on with @RadianceFields to discuss.
https://x.com/bilawalsidhu/status/2064146894930989138

holy crap! apple just beat google to the punch — 3d gaussian splatting is coming to apple maps. these 3d scenes are made from oblique aerial imagery. but unlike blobby photogrammetry — no more broccoli trees, no more melted powerlines — ground level detail that actually holds
https://x.com/bilawalsidhu/status/2064057313057439795

ICYMI google’s been shipping gaussian splats in maps for a while — but tucked inside immersive view and indoor only. I used it heavily in NYC. Meanwhile apple’s bringing radiance fields to 300+ cities this fall, built from aerial imagery. Surprised they didn’t drop ground-level
https://x.com/bilawalsidhu/status/2064187152930365502

China

BYD Robots: BYD announces humanoid robots with auto dealer distribution

Last week, BYD Executive VP Stella Li said that BYD is “”developing humanoid robots.”” The headline: BYD plans to sell them through its global auto dealer network if they reach homes. – Initial deployment: overseas dealerships, especially European vehicle stores, for demos,
https://x.com/TheHumanoidHub/status/2062437400522207604

Datacenter: China throws $295 billion down to build nationwide AI data centers

China Plans $295 Billion Investment to Build Nationwide AI Data Centers – Bloomberg
https://www.bloomberg.com/news/articles/2026-06-09/china-prepares-295-billion-plan-to-fund-nationwide-ai-buildout

Medicine: China approves world’s first commercial brain-computer implant, beating Neuralink

China beats Elon Musk to launch world’s first commercial brain chip
https://nypost.com/2026/06/07/tech/china-beats-elon-musk-to-launch-worlds-first-commercial-brain-chip/

Google

Gemma: Google DiffusionGemma: 4x speedier text diffusion model (different than next token prediction- worth knowing)

Awesome to see this innovation in text diffusion. DiffusionGemma is lightning fast, 4x faster than other Gemma 4 models! Congrats to @bodonoghue85 and the team who worked so hard on this – excited to see what people build with it!
https://x.com/demishassabis/status/2064873362799600042

DiffusionGemma is an open, experimental model that brings our text diffusion research to Gemma 4. It’s a racehorse 🏇achieving up to 4x faster inference by generating entire blocks of text simultaneously vs predicting token-by-token (word-by-word) output!
https://x.com/sundarpichai/status/2064744343743922189

DiffusionGemma is out 🔥 it’s compute-bound so 4x faster compared to other Gemma-4 models (1k tok/s on H100) 💨 also great on coding, generate and iterate on any code from 3D generation to front-end ⤵️
https://x.com/mervenoyann/status/2064753402064601181

Introducing DiffusionGemma
https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/

Search: Google continues to turn Search into an AI tool

🧵At #GoogleIO last month, we announced the next era of Google Search. Here’s what to know ⬇️ 1️⃣ We’re introducing a new, intelligent Search box, marking its biggest upgrade in over 25 years. Designed to anticipate your intent, it dynamically expands and helps you formulate your
https://x.com/Google/status/2064034586762354893

SpaceX: Google to rent 110,000 SpaceX GPUs for $920M/month

Google Taps SpaceX for $920M Per Month AI Compute Deal
https://winbuzzer.com/2026/06/06/google-taps-spacex-for-920m-monthly-ai-compute-deal-xcxwbn/

Translate: Google launches Gemini 3.5 Live Translate with support for 70 languages

Gemini 3.5 Live Translate is here
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-live-3-5-translate/

Anthropic

Business: Heavy sigh: Anthropic and OpenAI reportedly spend $1,000 to earn $100 in revenue

Anthropic/OpenAI may be spending more than $1000 for every $100 you pay them – R&A IT Strategy & Architecture
https://ea.rna.nl/2026/06/07/anthropic-openai-may-be-spending-more-than-1000-for-every-100-you-pay-them/

Kimi

Local agent 300 agents: Kimi Work launches local desktop agent running 300 parallel workers (what!)

Meet Kimi Work – a local AI agent on your desktop that does the work for you. 🔹Native agent swarm: Up to 300 AI agents running in parallel on your local machine. 🔹Browser use: Paired with WebBridge extension, your agent will navigate websites in your browser: search, scroll,
https://x.com/Kimi_Moonshot/status/2063990409903112344

Medicine

Mayo: Mayo Clinic AI spots pancreatic cancer three years before diagnosis

Mayo Clinic researchers have developed an AI model that detects pancreatic cancer on routine CT scans up to three years before clinical diagnosis. Published in Gut, the study tested the model, called REDMOD, on nearly 2,000 scans, including pre-diagnostic scans originally read
https://x.com/TheRundownAI/status/2064416920191869191

OpenAI

Charts: ChatGPT embeds chart creation and data tools on on mobile and web apps (so long, business intelligence jobs)

Turn data and comparisons into charts, directly in ChatGPT. Available now on mobile and web.
https://x.com/ChatGPTapp/status/2064018770839113769?s=20

Economic: OpenAI launches research exchange to study AI’s economic impacts

Introducing the OpenAI Economic Research Exchange | OpenAI
https://openai.com/index/economic-research-exchange/

Government: OpenAI in talks to give US government equity stake (OK, you have my attention)

Trump administration, OpenAI discussing possible government stake
https://www.cnbc.com/2026/06/05/trump-open-ai-altman-stake.html

Image search: OpenAI’s Responses API includes image results in web searches

Your app can now search the web for images. Web search in the Responses API now supports image results in addition to text results, so you can build apps that surface products, places, visual references, and source links for inspiration.
https://x.com/OpenAIDevs/status/2064395155688616153

Lockdown: OpenAI launches Lockdown Mode to curb prompt injection data theft (this is relaxing)

Lockdown Mode | OpenAI Help Center
https://help.openai.com/en/articles/20001061-lockdown-mode

Lockdown mode is now available in ChatGPT. We rolled this out for organizations a few months ago, and now it’s available for all users on all plans. Lockdown Mode is designed to help prevent the final stage of data exfiltration from a prompt injection attack by limiting
https://x.com/cryps1s/status/2062923575049531422

Memory: ChatGPT memory can retain context across chats (yes please!)

big upgrade to chatgpt memory rolling out today!
https://x.com/sama/status/2062660086787613116

much better ChatGPT memory:
https://x.com/gdb/status/2062608071411540196

We’ve been researching new ways for ChatGPT memory to carry context across conversations and keep it useful over time. Today, that work is rolling out as a more capable memory system in ChatGPT.
https://x.com/OpenAI/status/2062567556524003631

Plan: OpenAI attempts to be more obnoxious than Anthropic with boorish essays (but it’s a must read)

Built to benefit everyone: our plan | OpenAI
https://openai.com/index/built-to-benefit-everyone-our-plan/

Here is our current plan for OpenAI:
https://x.com/sama/status/2064088940932641225

Visa: ChatGPT adds Visa payments for autonomous AI shopping agents

Visa brings payments to ChatGPT as AI agents start buying for you | AP News
https://apnews.com/article/visa-chatgpt-openai-shopping-mastercard-d769dec86344cb4977c98789e8ec492f

Web apps: ChatGPT Codex launches Sites for building shareable web apps

build and publish web apps with chatgpt! i really wish i had this when i was a kid, but i do miss hypercard.
https://x.com/sama/status/2062661071761211561

Building apps has never been easier. With Sites, Codex can turn your work, ideas, and plans into an interactive website or app your team can explore, use, and share with a URL. Rolling out to Business and Enterprise plans, before expanding more broadly.
https://x.com/OpenAI/status/2061845949170045346

SpaceX

IPO: SpaceX IPO – All sorts of wild stuff in there.

Opinion: Here are all the red flags in the SpaceX IPO – MarketWatch
https://www.marketwatch.com/story/here-are-all-the-red-flags-in-the-spacex-ipo-7c55b31e

SpaceX IPO: 8 Things To Know Before It Goes Public
https://www.forbes.com/sites/investor-hub/article/spacex-ipo-things-to-know/

SpaceX’s IPO Filing And Some Interesting Points
https://www.forbes.com/sites/eriksherman/2026/05/23/spacexs-ipo-filing-and-some-interesting-points/

Watch @ElonMusk provide a technical update on SpaceX’s capability to manufacture, launch, and operate AI satellites at scale →
https://x.com/SpaceX/status/2064099405758906727?s=20

Anthropic

S1: Anthropic files IPO paperwork with SEC

Anthropic confidentially submits draft S-1 to the SEC \ Anthropic
https://www.anthropic.com/news/confidential-draft-s1-sec

Anthropic confidentially submits draft S-1 to the SEC | Hacker News
https://news.ycombinator.com/item?id=48358646

Science: Claude for Chemistry

Making Claude a chemist \ Anthropic
https://www.anthropic.com/research/making-claude-a-chemist

New Science Blog: Why has AI advanced faster in coding than in biology? To agents, bio databases are like cities built before cars—maddening to drive in because they’re designed for different traffic. How do we build infrastructure agents can use?
https://x.com/AnthropicAI/status/2064054837294354677

Apple

WWDC: Apple WWDC 2026 keynote (unreal that there are no great announcements)

Apple WWDC 2026 June 8: Introducing Siri AI and more – YouTube
https://www.youtube.com/watch?v=hF8swzNR1-o

Apple WWDC 2026 keynote in 25 minutes – YouTube
https://www.youtube.com/watch?v=2TEeQjoY05c

Siri: Apple launches Gemini-powered Siri AI (it can’t get worse)

Say hi to “”Siri AI””—Apple announces new, more “”conversational”” voice assistant – Ars Technica
https://arstechnica.com/apple/2026/06/say-hi-to-siri-ai-apple-announces-new-more-conversational-voice-assistant/

Business

Karpathy: Karpathy predicts AGI will barely register in GDP statistics – he’s one of the smartest folks out there

Andrej Karpathy thinks AGI’s impact on the economy will just be folded into the existing rate of growth. AI will be barely noticeable in GDP statistics. When he came on the show, I pushed back, saying AGI will cause a massive jump in productivity and growth. Watch our
https://x.com/dwarkesh_sp/status/2064754995086721170

Andreesen: Marc Andreessen says AI is not a bubble (VCs say VC things, I often reject his thoughts, but this is an OK essay)

Marc Andreessen introspects on The Death of the Browser, Pi + OpenClaw, and Why “”This Time Is Different
https://www.latent.space/p/pmarca

Debate: NYT roundtable debates AI’s future impact on workers – it’s many of the thoughts I’ve had and a good overview for laypeople

The New York Times published a roundtable discussion between @DAcemogluMIT, @deanwball, @clarashih & myself about the future of AI & who wins at work. I think it is a really nice overview of the core debates on the topic, and has some fun examples.
https://x.com/emollick/status/2064344218693259280

Google

Notebook: NotebookLM gets updates but I am not as interested as when it launched

Do your best research with NotebookLM
https://blog.google/innovation-and-ai/products/notebooklm/better-research-notebooklm/

Intel: Google and Nvidia weigh Intel as backup chip builder

Google and Nvidia Consider Intel as Backup Chip Manufacturer — The Information
https://www.theinformation.com/articles/google-nvidia-consider-intel-backup-chip-manufacturer

Studio: Google AI Studio (something I’ve almost never heard of) has built 18 million apps – what does this even mean

In @GoogleAIStudio we are now making more than 1,200,000 apps a week (and growing) with more than 18,000,000 created since late February 🤯 The progress continues!!!
https://x.com/OfficialLoganK/status/2064423388928790943

OpenAI

S1: OpenAI files confidential S-1, joins Anthropic and SpaceX toward IPO

Confidential submission of draft S-1 to the SEC | OpenAI
https://openai.com/index/openai-submits-confidential-s-1/

Exclusive: OpenAI Preps New AI Model, Expects To Go Public ‘Within the Next Year’ — The Information
https://www.theinformation.com/briefings/exclusive-openai-preps-new-ai-model-expects-go-public-within-next-year

OpenAI files confidential SEC S-1 paperwork for IPO | Fortune
https://fortune.com/2026/06/09/openai-files-confidential-s-1-sec-ipo/

Datacenter: OpenAI in talks for 10 GW Ohio site backed by Nvidia – Can’t fail, won’t fail… hashtag

OpenAI weighs Nvidia-backed lease for 10 GW Ohio data center campus | Network World
https://www.networkworld.com/article/4183513/openai-weighs-nvidia-backed-lease-for-10-gw-ohio-data-center-campus.html

Publishing

Scorcese: Art Directors Guild condemns Scorsese’s AI startup advisory partnership and endorsement

Art Directors Guild Slams Martin Scorsese for AI Partnership
https://variety.com/2026/film/news/art-directors-guild-statement-martin-scorsese-ai-1236770996/

First Aronofsky, then Scorsese, and now Gareth Edwards. Generative media has been polarizing – but is the Overton window shifting before our eyes?
https://x.com/bilawalsidhu/status/2062358867510440354

Runway: Speaking of movies, Lionsgate takes equity stake in Runway, launches joint IP program

Runway News | Runway and Lionsgate Expand Partnership
https://runwayml.com/news/runway-and-lionsgate-expand-partnership

Ad AI: New York law requires ads to disclose AI-generated performers (this is going to be the new “Photoshop” blurry line)

Synthetic performers’ in ads must be identified as AI as new New York law takes effect | AP News
https://apnews.com/article/new-york-ai-law-hochul-synthetic-performers-e433625bfb61c8abeab0d619869192ed

Sports

FIFA: FIFA kicks off (get it) faster offside tech and AI analysis at 2026 World Cup – Multimodal AI continues to be the most fun

Faster offside decisions, more stable referee body cams and more analysis opportunities for teams: how innovation is elevating the FIFA World Cup 2026™ experience
https://inside.fifa.com/innovation/news/offside-decisions-referee-body-cams-innovation-world-cup-2026