Tesla Roadster specs, Meta’s 816B model, and vibe coding at work
Tesla Roadster: production set for 2028, no price yet
Tesla unveiled the second-generation Roadster by launching it up a curved ramp into a controlled landing, then jumping it over five other Tesla vehicles mid-air. The specs: 0 to 60 mph in about one second, top speed over 230 mph. Production is scheduled for 2028. Tesla has not released pricing.
Meta Muse Spark Open: 816 billion parameters, runs on two Mac Studios
Meta released Muse Spark Open, which it positions as the top US open-weight model. The model has 816 billion parameters total and uses a mixture-of-experts architecture that activates 42 billion parameters per token. It supports a context window of up to 1 million tokens and accepts text, images, video, and audio as inputs, though it only outputs text.
Meta’s benchmark hardware was a pair of M5 Ultra Mac Studios with an unspecified memory configuration. That puts a frontier-class open-weight model within reach of well-equipped indie operators, at least in theory.
Companies building internal vibe-coding tools for non-engineers
Tech companies are quietly shipping internal vibe-coding apps for employees who do not write code professionally. The reported reasoning: internal tools get higher adoption than vendor products, build AI expertise in-house, and keep internal data away from third-party systems. The investment is small; the upside is custom tooling on your own timeline without waiting on a vendor’s roadmap.
⚛️ Google cuts quantum error rate 40x with magic state cultivation
Google Quantum AI demonstrated that a technique called magic state cultivation works in practice on its Willow processor. Magic states are needed to run operations that go beyond what classical computers can do. Creating them has historically been resource-heavy. Google’s new protocol achieved a 40-fold reduction in errors, which the team describes as a key step toward scaling quantum computing to practical use.
Claude is now reverse engineering hardware
Developers are using Claude to work through hardware debugging tasks that previously required reading thousands of lines of USB capture logs by hand. According to one write-up, Claude can spot byte-level sequence patterns, suggest what a changing byte likely means, and write tests to confirm its hypothesis. A human still decides what to build and validates the fix on real hardware, but the tedious middle work is shifting to the model. The post links to a harness with tools for reverse engineering Linux drivers.
CoreWeave debt hits distress signals as AI capex concerns grow
CoreWeave has taken on more than $51 billion in debt and obligations to fund its AI infrastructure buildout. The spread on its five-year credit default swaps has soared, and yields on its bonds continue to climb. Investors are not questioning whether demand for CoreWeave’s services is real. The concern is whether cash flow will arrive fast enough to justify the capital spending.
Also worth reading
- How Codex happened: The inside story of how OpenAI turned a cloud product without product-market fit into the centerpiece of its comeback.
- How to train your robot: Robot training data is expensive and scarce. Researchers are betting that the same scaling laws that transformed LLMs will eventually apply to robotic capabilities.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
