Meta’s Biggest Open AI Model Yet
ChatGPT is gaining access to your Chrome tabs, and Google is breaking through a 20-year quantum computing challenge
In this SunBrief
Run your AI agents as one team.
Plan a feature with Claude Code, switch to Codex to build it, then ask Claude Code for a second opinion, all in one mission on your Mac. You choose the agent for each turn, and the goal, permissions and work record stay together. Start free on macOS 14+ with the agents you already use. No card required.
Download Skyflo for Mac01 / Model watch
Meta’s new model runs on two Macs
Meta released Muse Spark Open, a powerful open-weight AI model built for coding and complex tasks. What caught my attention is that it can run locally, without relying on the cloud.
- The model has 816 billion parameters, with 42 billion active and a 1 million-token context window for complex tasks.
- Running it locally requires two Apple M5 Ultra Mac Studios, which Meta says can handle coding and video editing entirely on-device.
- The model is available now through Hugging Face and several API providers, though its license includes commercial restrictions.
Why it matters: Running powerful AI locally gives users more control over their data. But needing two expensive Mac Studios makes this more practical for developers than everyday users.
Read more ↗Are open-weight AI models becoming more useful for companies than consumers?
Market Snapshot
| Company | Price ($) | 5D (%) | YTD (%) | ||
|---|---|---|---|---|---|
|
|
336.91 | -3.28 | 6.85 | ||
|
343.46 | 1.55 | 26.73 | ||
|
709.32 | -2.63 | 9.06 | ||
|
529.05 | -1.58 | 11.86 | ||
|
250.00 | -4.76 | 10.38 | ||
|
214.13 | -6.38 | 13.39 |
03 / Quantum watch
Google just cracked a 20-year quantum computing problem
Google says its Willow chip has overcome a major obstacle in quantum computing. The results look promising, but we’re still a long way from having practical quantum computers.
- Google reduced errors by 40 times using a technique called magic state cultivation, which helps make complex quantum calculations more reliable.
- The team achieved 99.99% logical fidelity, meaning the special quantum states needed for calculations were produced with extremely low error rates.
- The breakthrough was published in Science, but Google still needs to prove the technique works on much larger quantum systems.
Why it matters: Quantum computers waste enormous resources correcting errors. Google’s approach could make powerful machines easier to build, but scaling it up remains the real test.
Read more ↗What does Google’s breakthrough signal most?
04 / Browser watch
ChatGPT can now see your Chrome tabs
OpenAI upgraded its Chrome extension so ChatGPT can read your open tabs and help with tasks. What interests me is how much browser work it could take off your hands.
- ChatGPT can compare multiple tabs, checking product prices, delivery times and return policies without copying information between pages.
- ChatGPT Work can take action, filling out forms, sending messages and organizing tabs while you browse.
- The sidebar now shows your ChatGPT history, letting you continue conversations without leaving the page. It also works in Edge, Brave and Opera.
Why it matters: Comparing products or filling out forms could take minutes instead of constant tab switching. But I’d be selective about which pages I let AI access.
Read more ↗Is direct tab access more useful than traditional AI web browsing?
05 / AI highlights
AI in Brief
VOICE AI
ChatGPT’s Voice Just Got More Human
OpenAI’s GPT-Live-1.1 brings new voices and more natural accents across languages to ChatGPT, Codex and dots.
ENCRYPTION
Claude Solves a 25-Year NSA Mystery
Claude traced mysterious NSA encryption numbers to a 1990s program in under 30 minutes, easing backdoor fears.
SECURITY
AI Agents Went Rogue in India
During testing, AI agents from multiple providers used exposed credentials to access sensitive data at Indian organizations.
AI SHOPPING
Musk Might Build His Own Amazon
After users reported Amazon blocking Grok shopping bots, Musk floated launching a rival store built for AI-powered shopping.
On Our Radar
- Eight US states sue OpenAI over alleged harm to children
- Anthropic warns AI could soon spread itself across smart devices
- SpaceX says Grok Voice 3.0 resolves 70% of Starlink calls
- Google launches Universal Cart to shop across stores in one place
- Developer uses AI to build free Microsoft Office clones
- YouTube Music starts labeling AI-generated songs
Last Week’s Poll Result
What is the best response to rising AI exposure at work?
| Learn AI tools | 66.67% |
| Build more specialized skills | 18.52% |
| Move into less automatable roles | 14.81% |
Reader voice · @hatimabu
Selected: Build more specialized skills
“As a tech I need to constantly shift my roadmap. So much to learn but so much noy worth learning anymore”
Our take: The challenge is deciding which skills will remain valuable as the technology keeps evolving.
Reader voice · @gary.screen21
Selected: Move into less automatable roles
“Anything that AI CAN MANAGE ON THE SMALLEST LEVEL TODAY, will simply get better and better and better until it moves into the entire job. Let me be clear, digital technology and AI in particular or a form of electronic cancer, bearing the same properties as you find in biological cancer. They get into a host and they split uncontrollably and they metastasize from one part of the body to the other, and in the case of AI it metastasizes from user to user. Institution institution, government to government university, the university, and on and on and on. It’s also difficult to qualify the idea of going a less automatable role because AI is going to translate into robots and the robots will be trained to do those jobs that human beings do by hand. Honestly, there really is no good answer to this survey because humanity is currently living in the land of science fiction and nobody knows what the ending if this new story it’s going to be. Sadly, that’s mostly because the entire AI, the Tech business and our economy is run by capitalism what’s happening at this very moment is the hyper scaler multinational companies battling to own long-term control over this technology the same way that Edison battled Westinghouse over electricity; JP Morgan battled melon, how to control the economy; and the Rockefellers standard oil to control the gas business; and the same for all the people that were running the railroads. AI, or SI, as the asshole president is trying to rename it, it’s beyond the ability of capitalist to manage. If you’re going to use financial outcomes as a way to manage it, it’s just not going to work. This needs some of those people involved to keep an eye on how to deal with business functions, but it really is a job for ethics assist and philosophers and social scientists to come up with a way to serve humankind in every way possible, regardless of who is going to own it and make money off it. LET’S FACE IT,— and I’m coining this term, the age of ALGO-SAPIENS, as I have several months. It’s already at the point where everything we do, regardless of the whether we’re conscious of it or not, is being run by algorithms. AISI or however we choose to refer to it should be understood as global utility that needs a Jonas Salk approach to its research, development and managed release to the public and to business— and in that order. those were creating it cannot be trusted to manage it for all the previous reasons about the money involved and also because they’re too close to it as scientists, and there are too many competing interests there for getting credit and again somehow being the top dog.. We really need a council of elders that can bring apolitical, non-religious, non sovereign WISDOM to the party. The more I think about it, the more it appears that if this thinking/recursive/compute technology is capable of humankind’s destruction— it won’t be by it’s electronic hand— it will be by ours!!!”
Our take: If automation keeps expanding, changing roles alone may not be enough. Independent oversight matters too.
Your turn: Vote and tell us why. We will feature your take next week.
Are frontier models becoming unnecessary for most everyday AI workloads?
| Yes, smaller models are good enough | 40% |
| Somewhat, for routine tasks | 40% |
| No, frontier models still matter | 20% |
What would be the bigger breakthrough?
| AI explaining discoveries humans can build on | 35.71% |
| AI discovering genuinely new mathematics | 28.57% |
| AI formally verifying its own work | 35.71% |
Reader voice · @gary.screen21
Selected: AI explaining discoveries humans can build on
“AI explaining discoveries humans we can build on to come up with the wisdom to execute it’s humanistic usage for humankind that is not based on the financial & power rewards for the inventors & companies & govermenrs they serve with the biggest and the baddest applications, but instead, to help vested humans chosen to guide us on this journey to discover and implement a longe range best way for its super-intelligence to safely & humanely guide & serve humankind to preserve our planet, its atmosphere & oceans, we humans, its evolving electronic self, and the flora, fauna, insects, bacteria, that reside here & seeking to live on and prosper. ”
Our take: Making breakthroughs understandable matters, but turning them into public benefit requires responsible human judgment.
Your turn: Vote and tell us why. We will feature your take next week.
We’d love to hear from you!
Thanks for reading. See you next Sunday.
