🤝 Apple and Meta Team Up to Run AI Locally

Apple and Meta have teamed up to run Meta’s new Muse Spark Open locally on Macs. It still takes serious hardware, but if local AI becomes practical for more people, cloud providers could have to work harder to keep their customers. You might never run a model yourself and still benefit from the competition it creates. Let’s dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Meta’s newly released Muse Spark Open can run locally on Mac Studios through a collaboration with Apple.

Key Points:

Meta and Apple worked together to bring the newly released Muse Spark Open to Mac Studios, using Apple’s MLX software to divide the model and its processing across two connected machines. Meta demonstrated what that setup could do through Muse Code, which fixed an iOS app and reviewed footage for editing in DaVinci Resolve while the model ran entirely on the Macs. Developers can download and adapt the model to build similar tools themselves, with support extending to compatible Windows and Linux systems that have enough memory to run it.

Apple can benefit from selling the computers that run Muse, while Meta gains from developers building with its model. Their collaboration could also give those developers more bargaining power as they build AI into their businesses. Depending on a hosted model means accepting that its provider can change prices or withdraw access, potentially forcing changes to software people already rely on. A capable model that developers can run themselves gives them another option when deciding where to build, although the hardware and upkeep remain substantial barriers. If that option becomes affordable and dependable enough, its influence could extend beyond the people who use it. Cloud providers would have a stronger reason to justify their prices and keep serving existing customers well when those customers have a practical way to leave.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning (constraints, edge cases, examples, tone) and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

Used by teams at Microsoft, Amazon, and Notion. Try Wispr Flow free on Mac, Windows, iPhone, and Android.

Start flowing free

Evolving AI: Anthropic is bringing independent outside experts into its labs to investigate AI risks and see whether its safety practices hold up.

Anthropic’s new arrangement gives METR reviewers a closer look at its internal safety work, building on a partnership that already includes reviews of several risk reports. It follows CEO Dario Amodei’s September pledge to bring independent evaluators into the lab, with employees encouraged to share information as they would with colleagues assessing the same risks. Anthropic can still require sensitive information to be removed before a report is published, though METR can publicly question whether those cuts were justified.

In an earlier review, a METR researcher found that some AI agent activity was going unmonitored, and Anthropic fixed the bug within 24 hours. Giving independent experts a closer look could uncover more such gaps and help the public see whether they are being fixed. Still, METR depends on Anthropic for access and does not decide which models the company releases. Its public findings could give customers and policymakers evidence to question a release while there is still time for Anthropic to rethink its plans. People should not have to wait for an AI failure to learn that someone had warned about the risk and the company went ahead anyway.

Evolving AI: Most engineering leaders are still making AI investment decisions on intuition.

Datadog's free AI Playbook gives you a practical framework to move from ad-hoc AI bets to evidence-driven systems, featuring priority plays, enabling capabilities, and a 90-day action checklist.

You'll get:

Free download →

Evolving AI: Anthropic added billions in sales last quarter and still disappointed some investors expecting a $2 trillion stock market debut.

Anthropic is preparing to go public as investors look more closely at the enormous cost of building and running AI. The company has committed at least $518 billion to computing and cloud infrastructure over a decade, leaving prospective shareholders to judge whether future sales can support that spending. Its listing is expected in November, and even the lower valuation under discussion would be about 55% above the $965 billion set in its last funding round.

Anthropic has already signed computing deals that could limit its ability to cut costs if growth slows further. About 80% of its infrastructure commitments cannot be cancelled or require payment regardless of use. Its prospectus also warns that many large customers could reduce or stop spending because they are not tied to long-term contracts. That gives investors a reason to look closely at how much customer spending Anthropic can reliably count on. Anyone buying its shares will need to judge how long the company could manage sales below its forecasts before having to raise more money to meet commitments it has already made.

QUICK HITS

🛒 Google launched its Universal Cart across Search and Gemini in the U.S., tracking prices and combining products from different stores.

🎙️ OpenAI began rolling out GPT-Live-1.1 in ChatGPT Voice, alongside improved accents and new voices in several languages.

🔬 CERN and Mistral partnered to build AI models trained on particle-collision data to help scientists make physics discoveries.

⚡ NVIDIA launched Nemotron 3.5 Super VL, a 120B-parameter open model it claims completes AI agent tasks up to 5× faster than rivals.

 *partner link