← Back to Home
April 10, 2026

Meta Goes Proprietary and Astronauts Brave 3000-Degree Reentry

Meta Abandons Open-Source Llama for New Proprietary AI Model Muse Spark
AI

Meta Abandons Open-Source Llama for New Proprietary AI Model Muse Spark

For the past two years, Meta's loudest argument in the AI wars was that open-source would win — that giving away Llama for free was both a noble cause and a clever competitive strategy. Apparently, they've changed their mind.

Meta has launched Muse Spark, its first proprietary AI model, marking a sharp and surprising departure from the open-source philosophy that defined its AI identity. Llama wasn't just a product — it was a positioning statement. Meta used it to differentiate itself from the locked-down, pay-to-play models coming out of OpenAI and Anthropic. Now, it's joining that same club.

This matters because Meta was one of the few major players genuinely keeping the open-source AI ecosystem alive at scale. Researchers, startups, and developers around the world built products and pipelines on top of Llama precisely because it was free to use, inspect, and modify. A proprietary pivot doesn't kill that ecosystem overnight, but it does raise real questions about who's going to carry the open-source torch going forward.

The timing is also worth noting. Meta is making this move as AI model costs remain punishingly high and the race to monetize generative AI is intensifying. Building a proprietary model gives Meta something it couldn't do with Llama — charge for it, control access to it, and build a moat around it. That's a very different business logic than "open source as strategy."

What we don't yet know is what Muse Spark actually does differently. Meta hasn't gone deep on benchmarks or specific capabilities, so it's hard to say whether this is a genuine technical leap or more of a strategic repositioning dressed up as a product launch. The name itself — Muse Spark — feels notably more consumer-facing than the utilitarian "Llama," which might hint at where Meta sees the model fitting into its broader product ecosystem, think Instagram, WhatsApp, Ray-Ban glasses.

There's also a philosophical tension here that Meta hasn't really addressed publicly. Mark Zuckerberg spent considerable time and energy arguing that open-source AI was safer and more democratic than proprietary alternatives. If the company is now building closed models, does that argument still hold? Or was open-source always more about competitive positioning against OpenAI than a genuine ideological commitment?

For the broader AI industry, this is a signal worth watching. If Meta — arguably the most credible big-tech champion of open-source AI — is moving toward proprietary models, it suggests the economics of openness may be getting harder to justify as models become more expensive to train and competition for enterprise contracts heats up. The open-source AI movement isn't dead, but it just lost one of its most prominent sponsors.
Source: VentureBeat
SPACE

Artemis II Crew Faces Nail-Biting 14-Minute Reentry After Historic Moon Mission

Four astronauts are about to slam into Earth's atmosphere at nearly 24,000 miles per hour, endure temperatures approaching 3,000 degrees Fahrenheit, go completely silent for six minutes, and then — if everything goes right — splash down in the Pacific Ocean like nothing happened. Welcome to the most dangerous 14 minutes in human spaceflight right now.

The Artemis II crew is returning to Earth Friday evening, with splashdown targeted for 8:07 pm ET off the coast of Southern California. The mission itself has been a landmark — the first time humans have traveled to the vicinity of the Moon in more than 50 years. But the homecoming, as is often the case in spaceflight, is where things get genuinely terrifying.

Here's the sequence that should be on your radar. About 44 minutes before splashdown, the Orion spacecraft's Crew Module separates from its Service Module — the European Space Agency-built section that has powered and propelled the vehicle for the past nine days. That separation exposes the heat shield for the first time, which needs to be angled with precision before the spacecraft hits the atmosphere. Flight director Jeff Radigan put it plainly: "We have to hit that angle correctly." There is no margin for ambiguity here.

At around 7:53 pm ET, Orion enters the upper atmosphere at 400,000 feet. Within roughly 24 seconds, the spacecraft is engulfed in plasma, and all communication with Mission Control cuts out for approximately six minutes. That blackout isn't a technical glitch — it's physics. The ionized gas surrounding the capsule simply blocks radio signals. For those six minutes, the crew is on their own, and everyone on the ground can only wait.

The heat shield situation adds another layer of tension. During the uncrewed Artemis I mission in 2022, the heat shield performed — but not without anomalies that fell well outside what engineers had modeled or expected. NASA has studied those results extensively, but the honest reality is that this will be the first time humans have trusted their lives to an Orion heat shield during an actual lunar-return reentry. That's a different kind of data point than a test flight.

If reentry goes as planned, the plasma burns away, the spacecraft decelerates rapidly, and a series of parachutes deploy to slow Orion to a survivable splashdown speed. Recovery ships are already positioned in the Pacific.

What makes this moment feel bigger than a routine mission end is the context. Artemis II is the bridge between Apollo-era nostalgia and whatever comes next in human deep space exploration. A successful splashdown doesn't just bring four people home — it validates the architecture, the heat shield, and the entire Orion program that NASA and its partners have spent over a decade and billions of dollars building. The stakes, in other words, are about as high as they get.
Source: Ars Technica

Enjoyed this?

Get stories like this delivered every Tuesday — free.