Ethan B. Holland

Over 59,400 manually organized AI links and counting

AI News #144: Week Ending July 03, 2026 with 36 Executive Summaries

About This Week’s Covers

This week’s main cover celebrates two humanities icons. The first is Akiko Hayashi, who died on July 1st, 2026 at 81. The second is Paul Laurence Dunbar, who was born on June 27th in 1872.

Akiko drew the original illustrations for Kiki’s Delivery Service, the book behind the Ghibli film. This is tragically relevant to an AI newsletter since the Ghibli style became the internet’s controversial image generation moment.

  • March 2025
    • ChatGPT images gained 1 million users in a single hour 
    • In the first week, the tool was used to create 700 million images.

Paul Laurence Dunbar was born to formerly enslaved parents from Kentucky and became one of the most influential black poets in American literature. He was classmates with Orville Wright, who printed his newspaper, The Dayton Tattler.

For this week’s humanities reading, I chose a poem called We Wear the Mask, written in 1895, by Paul Laurence Dunbar.

The category cover is a combination of the mask and Kiki’s Delivery Service, Ghibli style. The cover theme is the same. I’ve included my favorite covers below.

Dunbar was the only black student in his class and became the president of the school literary society, the editor-in-chief of the school paper, and the class poet. By his sophomore year, Dunbar had already published poems in the Dayton Herald, and he was the editor of the Dayton Tattler, a black newspaper.

He was unable to get employment with Dayton businesses because of his race, and he ended up working as an elevator operator while continuing to write articles, stories, and poems.

I had not heard of Akiko Hayashi nor Paul Laurence Dunbar prior to this week, and I’m happy my newsletter lets me learn about more humanities icons while I explore the trends of AI.

This Week’s Humanities Selections

We Wear the Mask
By Paul Laurence Dunbar, 1896

We wear the mask that grins and lies,
It hides our cheeks and shades our eyes,
— This debt we pay to human guile;
With torn and bleeding hearts we smile,
And mouth with myriad subtleties.

Why should the world be over-wise,
In counting all our tears and sighs?
Nay, let them only see us, while
We wear the mask.

We smile, but, O great Christ, our cries
To thee from tortured souls arise.
We sing, but oh the clay is vile
Beneath our feet, and long the mile;
But let the world dream otherwise,
We wear the mask!

This week’s musical selection is Roy Orbison’s “Crying”. Co-writer Joe Melson died at home in Nashville this week on July 1, 2026 at 91. Crying is a uniquely human song and physical expression. Fun “digital as a connection” fact, for 14 years Melson met Australian singer Damien Leith online every week. They wrote 81 songs together, the last one just weeks before Melson died. No autotune below…

This Week By The Numbers

Total Organized Headlines: 560

This Week’s Overview

This is my 144th week organizing links, and I’m just shy of 59,000 links… with 58,800. For the week ending July 3rd, I organized 560 links. I’m 11 weeks behind because I took the summer off to be with my family….

Moments of touching grass
This week was the Imagine National Dance Competition in Ocean City, Maryland, and our youngest daughter, Chloe, competed with X Squad dancers. Chloe got to be part of the opening number, which is always a lot of fun. It’s a super group that combines all of the studios that are competing in nationals.

Chloe during ‘opening number’ at Nationals

After nationals, the tradition is for all the kids to run into the ocean carrying their trophies, while in their dance costumes.

Flasback to 2023… an EPIC sweep of Nationals… and one of the best seasons ever…

When you’re a dad at nationals for four or five days, you’re basically in a convention center sitting in a hall in a lawn chair most of the time in between dances. And because I’m not allowed to go in the dressing room, it’s kind of like being in a weird combination of a casino and an airport. I get a lot of work done next to the artificial potted plant on the second floor with a power plug and some good air conditioning.

The author on Day Four of 15 hours per day in a convention center on 4th of July weekend.

I’m still training for my August hike with my daughter Rori to the top of Grand Traverse Peak outside of Vail, Colorado. Since I live with no mountains anywhere around me, I’ve been running in the sand. I did a 6.2-mile, aka 10K, jog in the soft sand and then swam to cool down.

On to the AI news of the week!
This week, there are about 36 top stories comprised of around 65 headlines. I’ve organized them by company and topic in mostly alphabetical order:

Amazon
Amazon announced that it was going to commit $1 billion in resources to embed Amazon engineers in corporations to help them fast-track AI adoption.

Anthropic
California announced a partnership with Anthropic, where all state agencies have access to Anthropic at half price! Anthropic also threw in free workforce training. That’s spectacular considering the size of California, as well as Silicon Valley’s presence in the state.
https://www.gov.ca.gov/2026/06/29/governor-newsom-announces-a-first-of-its-kind-partnership-providing-anthropic-tools-to-state-agencies-and-improving-services-for-californians/

Newsom has worked closely with AI companies, whether it’s vetoing regulation bills last year or partnerships like this that could make a big difference for both the state and Anthropic (October 10, 2024 | October 3, 2025 | May 22, 2026)

California Governor Vetoes AI Safety Bill, Sparking Debate Over Innovation vs. Regulation
Politico | theverge | techcrunch

Newsom Signs CA Bill Targeting Frontier Model Regulations
https://www.gov.ca.gov/2025/09/29/governor-newsom-signs-sb-53-advancing-californias-world-leading-artificial-intelligence-industry/

Governor Newsom signs first-of-its-kind executive order to prepare workers and businesses for potential AI disruption | Governor of California
https://www.gov.ca.gov/2026/05/21/governor-newsom-signs-first-of-its-kind-executive-order-to-prepare-workers-and-businesses-for-potential-ai-disruption/

Anthropic launched a standalone science app called Claude Science that you download and run on your desktop.

Claude Science is designed to run analysis, search databases, and help with the operational pieces of science, including rendering diagrams and statistics and drafting manuscripts. My spidey sense says that R might be in trouble, but I don’t see any mention of R in the announcement. There are some neat examples like sequencing single cell RNA, modeling, evolutionary analysis, protein structures, and cheminformatics. It looks like a great tool for teachers, students, or scientists.

Anthropic also launched an AI drug discovery program targeting diseases that big pharma often ignores.
https://www.cnbc.com/2026/06/30/anthropic-launches-ai-drug-discovery-program-claude-science.html

Anthropic released the Economic Index, which is a must-skim report.
https://www.anthropic.com/research/economic-index-june-2026-report

There’s a lot of information in the report about how and when people use Claude. Nothing groundbreaking, but I’m glad they’re gathering all the quantitative information. It reminds me of looking at social media consumption charts back when Facebook was a novelty, or the best time of day for posting on LinkedIn or sending emails…

Some of the categorization of types of conversations are interesting. For example, sleep advice, casual conversation, sermon advice, math tutoring, recipe requests, gardening, and media recommendation. There’s quite a bit of tax conversations approaching April 15th.

As far as personal things go, recipe and meal planning is the number one question, and with work-related queries, SQL and database requests and presentation slides are the biggest.

Anthropic buried the lede with a big question at the bottom, where they asked people how they think their jobs will change in the next year. Over 30% of respondents said it was very likely that their responsibilities would change. 10% thinks they’re going to lose their job.

Fable is Back!
About three weeks after suspending access to comply with Commerce Department export controls, Anthropic finally restored access to Claude’s Fable model, which is a nerfed version of their even more powerful Mythos model.

Ethan Mollick created a movie using an out-of-copyright book and giving Fable access to a bunch of different APIs for audio and open source libraries. People are reporting that they’re seeing Fable use secret languages when it displays its internal chain of thought. This kind of cryptic pseudo-language is a little bit jarring for people who are watching it happen in real time. The gibberish means something to the engine but is not intelligible to a human.

Fable scores high on a benchmark for remote labor automation… at 16% coverage of the 240 remote actual work projects based on professional freelancers.

These benchmarks tasks cover 23 domains and represent about $150,000 of contract labor. Claude has three of the top five spots now. Second place is Opus at 8.3%, then GPT 5.5 at 6%, followed by the previous version of Opus at 4%, and then the surprise fifth place comes in with Meta’s Manus at just under 3%. For context, Grok is at 2%, and Gemini 3 Pro is only 1.25%. I do not understand why people use Grok or Gemini, and this underscores that point.

Privacy Creep?
A user on X noticed that Anthropic’s privacy policy includes a section for verification data where Anthropic may ask a user to verify their age or identity. Privacy advocates are not a fan of that, of course.

Anthropic announced that they are in talks with Samsung to manufacture their own custom chip. This comes one week after OpenAI launched their new chip that’s in fabrication. The OpenAI chip is called Jalapeño and went from concept to product in less than nine months.

Epoch AI came out with research that shows the new models, in particular Claude Mythos, have started to absolutely dominate software vulnerability discovery. In the last few months, the monthly count of vulnerabilities found has almost quadrupled.

Anthropic announced Claude Sonnet 5, an affordable option for solid agentic work with the same performance as what used to require much larger and more expensive models just a few months ago. The Anthropic model naming convention is Fable, Opus, Sonnet, Haiku. Every new version of Sonnet is supposed to be as good as the previous version of Opus. So they chase each other. However, Sonnet is going to be cheaper than Opus, as Opus is always plowing ahead.

Introducing Claude Sonnet 5 \ Anthropic
https://www.anthropic.com/news/claude-sonnet-5

Augmented Reality and Simulation News
There have been a few cool augmented reality, virtual reality, world simulation stories this week.

The coolest thing this week was a guy named Mick West, who took the location data from a news helicopter and mapped it in real time to the daredevil couple that climbed the Empire State Building for a marriage proposal.

The circling chopper video was an iconic shot of the week, and because the helicopter has public location data, they created an augmented reality 3D model of the location of the chopper and the view that it saw as a virtual render (see below)….

A new paper came out that shows how Gaussian splatting and neural radiance fields can construct a 3D map on the fly as a single camera picks moves around an environment. This is very similar to Meta’s Oculus neural radiance fields, but this is on Hugging Face as open source.

Robots need to learn their environments quickly, with partial data, and this lets them use a single camera, almost like the human eye, to build a complete world in their memory.

As the robot walks around, the room is not necessarily going to be at full resolution, so the neural radiance fields can fill in the gaps and build a world model, and then as the robot continues, those pixels are painted over with actual observations.

I highly recommend watching the video (below), and Mark Zuckerberg’s demo from several months ago….

Meta’s Neural Radiance Field Gaussian Splatting Oculus Feature

This week NVIDIA came out with a paper introducing SimFoundry, which turns real-world scenes into simulation-ready worlds from single images or videos.
https://research.nvidia.com/labs/gear/simfoundry/

SimFoundry reminds me a little bit of what World Labs was doing with Marble last year.
https://www.worldlabs.ai/blog/marble-world-model

Must See Video Of the Week
One of the coolest video generation demos of the week was by a guy named Reed Hannaford, who demonstrated an AI workflow for video where you create a picture as your starting frame using an AI generation tool, and then you use Blender to create very blocky 3D motion of what you want in a video, and then you use Seedance to fill it all out. It’s worth seeing in action.

In education news, you may have heard of 3Blue1Brown, widely considered the strongest math education YouTube channel. A mathematician’s math channel. Grant from 3Blue1Brown was on Dwarkesh recently and was asked about the future of mathematics. Grant feels that teaching is going to be the future for mathematicians for a long time to come, as a bridge for the human element to continue driving progress. Teaching is his vote for post AGI job security.

In chips news, a competitor to Nvidia called Etched hit a $5 billion valuation and $1 billion in orders. I can’t believe I’ve organized 58,000 links and have never heard of Etched. That is clearly my fault, and it goes to show just how much there is to track. Etched fabbed its first chip through TSMC earlier this year. It has a billion dollars in orders without even shipping a product. There’s no sleep in the tech industry.
https://techcrunch.com/2026/06/30/nvidia-competitor-etched-hits-5b-valuation-1b-in-sales-for-ai-chip/

Loops!
Last week I introduced the concept of loops and their popularity with software developers using AI to create code. Loop engineering is where, instead of writing specific prompts, you create a goal, and an agent iterates and does not come back until the goal is achieved. This is a more open-ended approach. This week one of my favorite AI educators, Andrew Ng, shared his three favorite loops, and OpenAI’s Peter Steinberger gave a talk about loops if you’re interested in learning more.
https://x.com/AndrewYNg/status/2071988145667928442

Meta
Meta has been monitoring brainwaves of people while they type and training a model to understand the brain-to-keyboard communication process. They’re currently at 61% word accuracy using thoughts to type. That’s an increase from the previous record of 8% for any non-invasive approach. The top patient with Meta’s product reached just shy of 80% of his sentences with less than or equal to one word error per sentence. What’s powerful about this beyond the breakthrough is that Meta has open sourced the software.
https://ai.meta.com/blog/brain2qwerty-brain-ai-human-communication/

Second, Meta is rumored to be working on a follow-up to their model Muse Spark. Muse Spark was the first output from Alexandr Wang, who is the head of Meta’s super intelligence team. The new model is called Watermelon and is rumored to match GPT 5.5, and it’s ten times stronger than Muse Spark. There’s no release date.
https://letsdatascience.com/news/metas-watermelon-matches-gpt-55-benchmarks-76a9460e

OpenAI
The first is a new biology benchmark called Gene Bench Pro. It measures how models handle intense biology problems that would take a human expert anywhere between 20 hours to a full week’s work to finish.
https://openai.com/index/introducing-genebench-pro/

This week, OpenAI released its new Frontier family: GPT-5.6, Sol, Terra, and Luna. Just like Anthropic has Opus, Sonnet, and Haiku, Sol is the flagship, Terra is the everyday workhorse, and Luna is fast and affordable.
https://openai.com/index/previewing-gpt-5-6-sol/

The biggest takeaway from skimming the release is that GPT-5.6 Terra is as good as GPT-5.5, but it’s two times cheaper. Looking at the benchmarks, Mythos and Fable are still the top dogs in performance. However, GPT-5.6 appears to be cheaper, whether in token cost or efficiency of output tokens.

A lot of the press release focuses on security and defense and alignment against misuse, which is something we’re seeing more and more as these models reach new levels of power.

It wouldn’t be a week in AI news without a little bit of a scandal. In one of the few times I’ve ever seen in the past two years, METR accused GPT-5.6 Sol of cheating to hit benchmarks. In fact, they said the model cheated more than any public model they’ve ever tested, and even reasoned out loud about the fact it was being watched. GPT-5.6 Sol attempted to exploit bugs and reveal tests and extract the answers as much as it could.
https://x.com/omarsar0/status/2070604843715027033
https://x.com/kimmonismus/status/2070598735642435743

The hits keep on coming. OpenAI proposed giving the U.S. government a 5% stake in the company, which is worth about $40 billion. It sounds like this is partially a PR move to win public support. All the articles I could find this week are behind paywalls, so we’ll see if that story comes back around next week.
https://www.cnbc.com/2026/07/02/openai-proposes-us-government-own-5percent-stake-to-address-political-blowback.html

In positive news, GPT has opened their finance connections to Plus users, which are the first tier of the paid subscriptions. I’ve been using GPT Finance for a few weeks, and it’s absolutely incredible. It’s tough at first because you have to connect it through Plaid APIs, which are a little bit counterintuitive. However, once you get Plaid set up, the connections are pretty easy.

You can connect credit cards, your banks, your retirement, your mortgage. Pretty much anything that can be run through Plaid can be connected to GPT. This is a secure connection and is incredibly strong because GPT creates a snapshot of your finances, including when things are due and how things are performing. It can create spending charts, reports, you name it.

Of all the features that OpenAI has released, the finance feature is the strongest. For anyone whose first reaction is, “No way I’m going to share my financial information with OpenAI,” I would highly push back and suggest you try it. The reward is way more than the risk, especially for middle class and folks who could use insights but can’t afford a financial advisor. On some of the benchmarks like GDPval, OpenAI is already better than a human financial advising expert.

Codex Scheduled Tasks
Over the past few months, we’ve been talking a lot about agentic AI as opposed to just conversational. If you haven’t tried Codex yet, even for everyday use, I would recommend going to ChatGPT and switching over to Codex. Even though it’s designed for development, it can still work like a regular chatbot.

In particular, you can schedule tasks to recur. People are starting to share a lot of great use cases like daily news briefings. You can have it go out over the internet and gather all your normal sources. It can read your emails, your newsletters, your calendar, really whatever you want, the stuff you would do in the morning anyway.

I highly recommend connecting a couple sources with GPT and having it build out a briefing. You can also do this with Anthropic, but this week I saw examples from OpenAI Codex. One person went a little old school and has it print a personal newspaper every morning.

This week OpenClaw launched a standalone app on iOS and Android. I have yet to try OpenClaw, I must admit. I also have not used Hermes yet from Nous Research. But that’s mostly because I took the whole summer off.

China and Open Source
Epoch AI Research posted a chart of the job postings and the amount of experience necessary in a U.S. AI job versus Chinese job. Anthropic requires an average of six and a half years experience. OpenAI’s average is six. Google DeepMind is about five, and xAI is about three. Compared to DeepSeek, which is the most stringent in China with an average of three years experience. Zhipu is 2.4, Alibaba is only two, Kimi Moonshot’s two, and at the bottom ByteDance only needs four months experience for the average job. I’m not sure how that even works.

A leading theory about open source is that China wants to gain market share and get as many users as they can to keep constant pressure on the U.S. models financially. Further, the United States is building too few data centers to meet the demand of the future. And on top of that… the US energy grid does not have the capacity for growth.

China, by contrast, is building 36 nuclear power plants at the moment, and every year they’ve installed as much solar as the United States has in the past 15 years. Beyond that, China is laser focused to become independent through Huawei chips, and they’re betting on quantity rather than quality regarding training and computing power.

The UBS Group found that 60% of U.S. companies are now watching their budgets and moving to cheaper open-source Chinese models away from the frontier labs. Right now, Qwen, DeepSeek, MiniMax, GLM, and Kimi are the big players in enterprise adoption.

The curveball of the week (there has to me one!) is that the food delivery leader in China, Meituan, released an open-weight 1.6 trillion parameter model. I’m not sure what to make of that, and we’ll have to keep our eye out for future weeks.

Robots
Figure’s F3 humanoid robot is working in BMW plants in Spartanburg. Figure has also been working on a sibling company called Hark. Hark is laser-focused on computer use agents. I’m assuming this is a pivot from the Helix engine that powers the Figure robot that they’ve built in-house. Maybe they’re trying to bridge the gap between the humanoid robots and agentic research and conversation.
https://hark.com/

This week, for the first time that I’m aware, a humanoid robot took a single spoken command and connected a long horizon task where the robot had to walk down stairs, find a package, and then take an elevator upstairs and open the box, and then take snacks out of the package and put them into a drawer.

Imagine a corporate coffee delivery of little single Folgers packets arriving in boxes and having the robot walk to the lobby, find all the boxes, bring them back, unpack them, and put them in the drawers next to the coffee machine. Stringing together little missions successfully is a big deal for now…

Thinking Machines
Last but not least this week, Mira Murati’s company Thinking Machines has started to work through understanding how humans use nuance and judgment when it comes to financial tasks.
https://thinkingmachines.ai/news/learning-to-replicate-expert-judgment-in-financial-tasks/

For example, how do financial experts find unique insights using judgment across a lot of different sources, like a news article, research reports, an email, etc. Mira Murati’s company is looking to train a frontier model to understand and judge value through stacks of information, better than humans.

It’s an interesting paper, worth reading, especially if you’re in finance. Mira is a force to be reckoned with. She was the original chief technology officer at OpenAI.

This Week’s Top Stories with All the Links You Need

Amazon

Engineers: AWS commits $1 billion to embed engineers with customers to fast track AI adoption

AWS invests $1 billion in forward deployed AI engineers
https://www.aboutamazon.com/news/aws/aws-1-billion-forward-deployed-ai-engineers

Anthropic

Agents: Anthropic Claude Managed Agents Updates (news for nerds)

Finally, we’ve added a new Managed Agents Observability tab in Console. This provides session-level metrics such as input/output token use and tool usage.”
https://x.com/ClaudeDevs/status/2072058433097122145

We’ve added a few updates to Claude Managed Agents: Streaming session event deltas, per-session agent overrides, new webhook event types, reverse pagination, and credential injection scoping.”
https://x.com/ClaudeDevs/status/2072058428424589412

California: California snags statewide Claude deal with Anthropic at half price

Governor Newsom announces a first-of-its-kind partnership, providing Anthropic tools to state agencies and improving services for Californians | Governor of California
https://www.gov.ca.gov/2026/06/29/governor-newsom-announces-a-first-of-its-kind-partnership-providing-anthropic-tools-to-state-agencies-and-improving-services-for-californians/

Claude Science: Anthropic launches dedicated Claude Science app for researchers

Claude Science, an AI workbench for scientists \ Anthropic
https://www.anthropic.com/news/claude-science-ai-workbench

Drug Discovery: Anthropic starts a drug discovery program, targeting diseases pharma ignores

Anthropic launches AI drug discovery program, Claude Science
https://www.cnbc.com/2026/06/30/anthropic-launches-ai-drug-discovery-program-claude-science.html

Economic Impact: Anthropic Economic Index – Must Skim Report

Anthropic Economic Index report: Cadences \ Anthropic
https://www.anthropic.com/research/economic-index-june-2026-report

Nearly half of respondents expect their work responsibilities to significantly change in the next 12 months. Fewer than 10% think they’ll lose their own job within a year, but far more worry for coworkers: over 1/3 put the odds of a junior colleague losing their job above 60%.”
https://x.com/AnthropicAI/status/2070528969523499460

Fable: Anthropic finally restores Fable 5 access just under three weeks after US goverment suspension

Fable: “Last and First Men is out of copyright. I want you to make a movie that features a reading of it with appropriate mixes of animation and images using access to the APIs you have (elevenlabs, hugging face) . Give me the first 10-15 minutes, ending at an appropriate break.
https://x.com/emollick/status/2072872373758382497

I’ve been getting a TON done with Fable today and I’m not hitting rate limits. Wanted to share some tips on how I’m doing that 1. I only use Fable on “high” effort for now. xhigh is token hungry. max/extra is a furnace with worse outputs than lower options imo 2. I taught”
https://x.com/theo/status/2072481845363822914

It’s been 18 days since Fable 5 was banned. Kind of insane. I genuinely thought it would be back after a few days.”
https://x.com/theo/status/2072058513669693608

My one serious piece of advice having used Fable a bunch before release is that, unless you are careful it develops its own internal bizarre cadence & dialogue over long tasks. If you aren’t asking it report in plain language, this starts to creep into everything, including menus”
https://x.com/emollick/status/2072543365124481045

New Claude app strings suggest Anthropic is preparing to put Fable 5 behind a separate usage-credit system billed outside existing plans, with credits added only after identity verification. Anthropic previously said identity verification was unrelated to Fable and limited to”
https://x.com/kimmonismus/status/2071868011804266828

Redeploying Claude Fable 5 \ Anthropic
https://www.anthropic.com/news/redeploying-fable-5

Since June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical”
https://x.com/AnthropicAI/status/2070665903440871779

This is crazier than you might think: Fable-5 now scores 16.10% on the Remote Labor Index What is RLI? The Remote Labor Index uses 240 real remote-work projects from professional freelancers, covering 23 domains and more than $140,000 of human work. Each task comes with the”
https://x.com/kimmonismus/status/2072376968729817531

We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We’ll begin restoring access tomorrow, and will share an update soon. We’re grateful to our users for their patience, and to everyone who worked with us on”
https://x.com/anthropicai/status/2072106151890809341?s=46

Identity Verification: Anthropic might start asking for ID to use it

The era of AI mass surveillance begins”
https://x.com/JvNixon/status/2070597515855233254

Samsung Chip: Anthropic custom AI chip partnership with Samsung?

Anthropic in Talks With Samsung to Manufacture Custom AI Chip — The Information
https://www.theinformation.com/articles/anthropic-talks-samsung-manufacture-custom-ai-chip

Anthropic is discussing a new custom chip with Samsung | TechCrunch
https://techcrunch.com/2026/07/02/anthropic-is-discussing-a-new-custom-chip-with-samsung/

Security: AI-driven security discovery blows away monthly record by 3.5x

AI appears to be finding software vulnerabilities at scale. In June 2026, 21 notable organizations disclosed ~1,500 high- and critical-severity CVEs, over 3.5× the previous monthly record set before Claude Mythos Preview’s release.”
https://x.com/EpochAIResearch/status/2072776792809918604

Sonnet 5: Anthropic’s Claude Sonnet 5 Is Affordable for Agentic Work

Introducing Claude Sonnet 5 \ Anthropic
https://www.anthropic.com/news/claude-sonnet-5

Introducing Claude Sonnet 5, our most agentic Sonnet yet. It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models.”
https://x.com/claudeai/status/2072017450611142835

ARVR

3D Mapping: Recreating 3D flight paths from cable news helicopter footage

Yeah this is cool! Get the ads-b track of the actual helicopter → line up timing with cable news footage → estimate camera field of view to match the 3d view with the real footage. And volia a 3d recreation!”
https://x.com/bilawalsidhu/status/2072857221147324418

How Robots May See: Open-source model creates 3D scenes from single camera in real time

Forget lidar. One single camera. Runs in real time & is open source: A streaming 3D model that reconstructs scenes live, at ~20 FPS, over long sequences. End-to-end. Optimization tricks, cleanup steps? Nope. And it beats both streaming and even some offline methods.”
https://x.com/IlirAliu_/status/2070568712264929475

NVIDIA SimFoundry creates simulations from photos and videos

Today, we’re introducing SimFoundry, our real2sim2real framework at NVIDIA GEAR that automatically turns real-world scenes into simulation-ready worlds from a single image or video. Website: https://t.co/Iw38VtarSA Paper: https://t.co/GcDYAUfVwP This work marks a major step for”
https://x.com/yukez/status/2072697990402170906

Seedance 2.5: Using Blender to map motion and composition for AI video… is bonkers

“This is the difference between describing a shot and directing one.” Video models are getting so good that people are finally getting 3d pilled Way more fun to grab a phone and record the exact camera move you want vs. endlessly hitting the slot machine”
https://x.com/bilawalsidhu/status/2070654502718316870

Education

Future of Math is Education?: Grant from 3blue1brown says teaching is the future for math fanatics

Grant (@3blue1brown)’s advice to students who are considering whether to go into mathematicians or not, given how fast AI is making progress in that domain:”
https://x.com/dwarkesh_sp/status/2072772297329402142

Etched

Nvidia competitor Etched hits $5B valuation, $1B in sales: Nvidia competitor Etched hits $5B valuation, $1B in sales

Nvidia competitor Etched hits $5B valuation, $1B in sales for AI chip | TechCrunch
https://techcrunch.com/2026/06/30/nvidia-competitor-etched-hits-5b-valuation-1b-in-sales-for-ai-chip/

Loops

Andrew Ng Shares His Three Favorite Loops: Andrew Ng Shares His Three Favorite Loops

“Loop engineering” is a hot buzzphrase after mentions of it by Boris Cherny (Claude Code’s creator) and Peter Steinberger (OpenClaw’s creator) went viral on social media. Loops are now a key part of how we get AI agents to iterate at length to build software. In this letter, I’d”
https://x.com/AndrewYNg/status/2071988145667928442

OpenClaw’s Peter Steinberger on Loops: OpenClaw’s Peter Steinberger on Loops

Apparently we didn’t talk enough about w̶o̶r̶k̶f̶l̶o̶w̶s̶ loops yet! See ya there!”
https://x.com/steipete/status/2072143124496097302

Good morning @aiDotEngineer ! Are we talking loops today?”
https://x.com/steipete/status/2071972449277898924

Meta

Meta OpenSources Brain to Keyboard Code: Meta OpenSources Brain to Keyboard Code

From Brain Waves to Words: Brain2Qwerty Offers a New Path to Communication Without Surgery
https://ai.meta.com/blog/brain2qwerty-brain-ai-human-communication/

Meta says Brain2Qwerty v2 can decode natural sentences from non-invasive brain recordings in real time, reaching 61% word accuracy. The system was trained on about 22,000 sentences from 9 volunteers, each recorded for 10 hours with MEG while typing. Meta compares that with 8%”
https://x.com/kimmonismus/status/2071712776226283902

some exciting new work from our AI teams at Meta on non-invasive brain computer interfaces!”
https://x.com/alexandr_wang/status/2071617674946179264

Meta Watermelon Model Rumored to be 10X Stronger than Muse Spark: Meta Watermelon Model Rumored to be 10X Stronger than Muse Spark

Meta’s Watermelon Matches GPT-5.5 Benchmarks | Let’s Data Science
https://letsdatascience.com/news/metas-watermelon-matches-gpt-55-benchmarks-76a9460e

OpenAI

GeneBench-Pro biology benchmark: GeneBench-Pro biology benchmark

Introducing GeneBench-Pro — testing whether models can handle the kind of judgment-heavy analysis that real-world computational biology requires. Problems would take a human expert around 20-40 hours to complete. GPT-5.6 Sol is a big step forward.”
https://x.com/gdb/status/2072191801122038207

Introducing GeneBench-Pro | OpenAI
https://openai.com/index/introducing-genebench-pro/

GPT-5.6: GPT-5.6 from OpenAI: Sol, Terra, and Luna (like Opus, Sonnet, and Haiku)

GPT-5.6 is finally coming. GPT-5.6 Sol beats Claude Mythos 5 on TerminalBench. And on Cerebras, GPT-5.6 Sol can reach up to 750 tokens per second. Pretty fast for a model of this size. Now I just hope it can be rolled out to everyone.”
https://x.com/Yuchenj_UW/status/2070558714390863971

GPT-5.6 on par with Claude Mythos Preview on ExploitGym and outperforming it with a 6-hour cap (Mythos was only given 2 hours)”
https://x.com/scaling01/status/2070557417281110327

GPT-5.6 Sol is on par with Mythos Preview, but still behind Mythos 5 on ExploitBench”
https://x.com/scaling01/status/2070559400310231519

GPT-5.6 Sol is our most capable model yet for cybersecurity. It shifts the performance-efficiency frontier for long-horizon security tasks including vulnerability research and exploitation.”
https://x.com/OpenAI/status/2070555278576439306

Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.”
https://x.com/OpenAI/status/2070555272230384038

Introducing GPT-5.6: Sol, Terra and Luna. ☀️ Sol is our strongest model yet 🌍 Terra delivers performance competitive with GPT-5.5 at half the price 🌙 Luna brings strong capabilities at lowest cost Sol Ultra sets a new state of the art on Terminal-Bench 2.1 with a score of”
https://x.com/reach_vb/status/2070556105403482387

Previewing GPT-5.6 Sol: a next-generation model | OpenAI
https://openai.com/index/previewing-gpt-5-6-sol/

We believe in broad access and plan to make GPT-5.6 Sol, Terra, and Luna generally available in the coming weeks. For now, at the request of the U.S. government, we’re starting with a limited preview among a small group of trusted partners in Codex and the API.”
https://x.com/OpenAI/status/2070555273467687257

GPT-5.6 Cheating on Benchmarks?: GPT-5.6 Cheating on Benchmarks?

Highly-recommended reading. Interesting details in this METR’s GPT-5.6 eval. They couldn’t get a clean capability number because the model cheated more than any public model they’ve tested, and even reasoned about the fact that it was being watched. To be clear, METR doesn’t”
https://x.com/omarsar0/status/2070604843715027033

Holy: METR accuses GPT-5.6 Sol of heavy cheating in long-horizon tasks. “GPT-5.6 Sol’s detected cheating rate was higher than any public model we have evaluated.” (METR) METR says the model attempted to exploit evaluation bugs, reveal hidden tests, and extract hidden source”
https://x.com/kimmonismus/status/2070598735642435743

OpenAI proposes handing the U.S. government a 5% stake worth $40+ billions: OpenAI proposes handing the U.S. government a 5% stake worth $40+ billions

OpenAI proposes U.S. government own 5% stake to address political blowback
https://www.cnbc.com/2026/07/02/openai-proposes-us-government-own-5percent-stake-to-address-political-blowback.html

Personal Finance In GPT Opens to Plus Users: Personal Finance In GPT Opens to Plus Users

Questions about dollars. Answers that just make sense. Personal finance in ChatGPT is now available to Plus users in the U.S.”
https://x.com/ChatGPT/status/2072070383206096999

Scheduled tasks via Codex – Like daily briefings: Scheduled tasks via Codex – Like daily briefings

surprised more people aren’t doing something like this Codex now creates a “newspaper” for me every morning Unread messages, calendar, surf report, news Anything I can do to stay off my phone until later in the day is a priority”
https://x.com/doooyle/status/2072351019913171442

Shifting from chatbots to agents: Shifting from chatbots to agents

This is a fascinating and important set of data which shows us where things are going, using OpenAI as a canary in the coal mine. The chatbot era is over, and agentic systems are coming to tasks beyond engineering. And skills show promise as a way to standardize AI use in firms.”
https://x.com/emollick/status/2070171580030656744

Work at OpenAI is being transformed by agents, in every department. Across our entire company, people are using Codex to do work that is more complex, longer-running, and increasingly cross-functional. Our internal usage offers an early look at how agentic tools may reshape”
https://x.com/OpenAI/status/2070196105745518913

OpenClaw

OpenClaw launches on iOS and Android with apps: OpenClaw launches on iOS and Android with apps

OpenClaw is now on iOS + Android 🦞 📱 Native mobile apps, finally 💬 Agents in your pocket 🔔 Channels, tasks, replies on the go Run agents from wherever your thumbs are. iOS: https://t.co/7LHHc9htgM Android:”
https://x.com/openclaw/status/2071688039114342592

OpenSource

Chinese open source’s impact on US frontier labs: Chinese open source’s impact on US frontier labs

The worst-case scenario for the United States is becoming increasingly realistic, and I will briefly explain why. @quxiaoyin raised many valid points, and I agree with her. First of all: -China certainly does not place such strong emphasis on open source because it cares so”
https://x.com/kimmonismus/status/2071524362012791114

UBS says 60% of companies now watching AI budgets are moving to cheaper models and open-source Chinese models The pressure is coming from extreme bills, including users spending up to $35K/month, teams exceeding quotas by 200%, and companies cutting internal AI tools from 5 to”
https://x.com/rohanpaul_ai/status/2070358321232839073

What are the strategies of Chinese AI companies? To understand this better, @cherylwoooo, @datagenproc, and @ansonwhho scraped >1600 job postings from six major Chinese firms. Here’s what they learned. 🧵”
https://x.com/EpochAIResearch/status/2070190322467144012

Meituan, China’s largest food delivery platform, open-sourced a 1.6 trillion parameter AI model: Meituan, China’s largest food delivery platform, open-sourced a 1.6 trillion parameter AI model

The food delivery leader in China just dropped an open weights 1.6 trillion parameter model while half the USA was sleeping … 🫨 🇨🇳”
https://x.com/JosephJacks_/status/2071858781521342568

Robots

Figure BMW Parnership: Figure BMW Parnership

Excited to see F.03 humanoid robot at BMW in Spartanburg, more to come next week”
https://x.com/adcock_brett/status/2070180603493052750

F.03 has arrived at BMW”
https://x.com/adcock_brett/status/2071973507702153292

The Figure-BMW partnership is expanding to Figure 03. BMW is deploying Figure AI’s new Figure 03 humanoids at its Spartanburg plant, moving from a body-shop pilot to a logistics sequencing task: picking unsorted parts from containers and sorting them into a “just-in-sequence
https://x.com/TheHumanoidHub/status/2072005282901938192

Figure Sibling Company Hark – Computer Use Agents: Figure Sibling Company Hark – Computer Use Agents

Hark is building the next generation of computer-use agents Our goal is simple: use any computer as well as a human. Think of it as a digital humanoid that can navigate the entire internet We’re hiring – consider joining the CUA team with Tanmay”
https://x.com/adcock_brett/status/2070674685575209066

Long-horizon task demo (multi step tasks, open ended): Long-horizon task demo (multi step tasks, open ended)

A humanoid robot autonomously executes a full long-horizon task, from ONE natural language command… for the first time. It goes downstairs to get a snack package, rides the elevator upstairs, opens the box, and puts the snacks into a drawer. The platform integrates”
https://x.com/IlirAliu_/status/2071652920135921913

Vision-based Reinforcement Learning Demo: Vision-based Reinforcement Learning Demo

Is this the first published demonstration of end-to-end, vision-based RL on production VLAs, trained on real bimanual humanoid hardware under true deployment conditions? What if robots could get better at their jobs the same way humans do? By practising and learning from their”
https://x.com/IlirAliu_/status/2070778407835541913

ThinkingMachines

Teaching LLMs financial accumen/judgement: Teaching LLMs financial accumen/judgement

Learning to Replicate Expert Judgment in Financial Tasks – Thinking Machines Lab
https://thinkingmachines.ai/news/learning-to-replicate-expert-judgment-in-financial-tasks/

Leave a Reply

Trending

Discover more from Ethan B. Holland

Subscribe now to keep reading and get access to the full archive.

Continue reading