Welcome back! Tired? Backflipping after shaking the dean’s hand at graduation. Wired, apparently? Booing anyone who gets too excited about AI. Former Google CEO Eric Schmidt got an earful speaking at the University of Arizona. And at one Florida ceremony, a speaker’s line about “the next industrial revolution” was met with boos.
Microsoft’s take on the trend: Students are “telling us what we need to hear,” president Brad Smith said this week. Curious what you think that really is—hit reply and share your thoughts on the next gen’s beef with AI tech.
Why Would Anthropic Release a Purposely Throttled AI Model?
Anthropic just dropped its most powerful public model ever, and it's already causing a major divide among users.
The new model is called Claude Fable 5, and on benchmarks it's a beast. Anthropic says it shows "exceptional performance" across software engineering, knowledge work, and vision. Some users are calling it the most capable AI they've ever used.
But others are running straight into a wall.
Here's why:
Anthropic also released a new version of its frontier model, Mythos 5, but only to a limited group of cybersecurity researchers, government partners, and trusted organizations through a program called Project Glasswing.
Fable 5 is built on the same underlying model as Mythos (same raw capabilities) but Anthropic has wrapped it in some of the strictest safety filters any major lab has shipped.
The problem? In some cases, if Fable detects prompts related to cybersecurity, biology, chemistry, model distillation, or frontier AI development, it either falls back to a weaker model or limits the quality of its response.
Critics argue that Anthropic is creating a future where the most powerful AI capabilities are only available to a small group of approved institutions while everyone else gets a filtered version.
On the other hand, Anthropic's argument is that the strongest models are now powerful enough that unrestricted access could create real or even irreversible risks.
My POV: What's interesting to me is that both sides are making reasonable arguments.
It's reasonable to want guardrails on something genuinely dangerous. It's also reasonable to be uncomfortable with a precedent of "the best AI is reserved for approved organizations." This is going to be one of the central tensions of frontier AI for the next few years, and we're watching it play out in real time.
— Matt
SpaceX's $1.75 Trillion Bet on Who Controls AI's Compute
Via Reuters
SpaceX begins trading on the stock market today, raising $75 billion at a $1.75 trillion valuation—the largest stock offering ever. Investors can barely contain themselves, but underneath the rocket emojis is a bigger story: SpaceX is betting that the next great compute platform won't be on Earth.
The tell is who's already paying Musk. In the last two months, SpaceX signed compute deals with Anthropic and Google worth roughly $26 billion a year combined, inherited from the xAI merger and its Colossus data center in Memphis. That second name matters—Google is one of the largest owners of AI compute in the world, and it’s still renting capacity from Musk to meet demand for Gemini.
But compute is going scarce: This is largely due to power, water, land, and local opposition. And Musk's answer is to move compute into orbit, where solar is constant and cooling comes free. This “moonshot” has three parts that have to be pulled off at once, as TechCrunch put it:
Fully reusable Starship launches: still unproven, with the most recent booster failing a controlled reentry
A new US chip foundry called Terafab: the kind of project that typically costs billions and takes a decade
A satellite factory capable of producing roughly 556 AI satellites a month: about double Starlink's current rate, in a plant that doesn't exist yet
The skeptics aren’t exactly quiet: Two independent SpaceX valuations this week landed far below the bankers' figure, with Morningstar at about $780 billion and NYU's Aswath Damodaran at $1.2 trillion. Morningstar frames the gap as essentially a $72-per-share bet on SpaceX delivering orbital data centers at the pace Musk promises.
Why it matters: Soon public investors will scramble for a piece of it all—a near-monopoly on access to space, a global internet network, and a wager on the most ambitious infrastructure project of the AI era. But what Musk is promising now may prove just as hard to reach as Mars.
Order Your DoorDash by Describing It
DoorDash launched an AI chatbot on Thursday called "Ask DoorDash," letting you order food and groceries with plain-language prompts and photos. The pitch: Stop scrolling through endless restaurants and just tell the app what you want. Finally, AI is coming for the real bottleneck in food delivery (AKA figuring out what everyone actually wants for dinner).
How it works: You can describe a craving ("a filling dinner for a family of four"), share a recipe link to auto-build a grocery cart, or snap a photo of your handwritten list. It'll add items in the right quantities—and even nudge you to check whether you already have staples like butter before buying more.
The bigger picture: DoorDash is joining a crowded race for agent food delivery. Uber Eats rolled out a "Cart Assistant" in February, and Instacart has its own shopping assistant. It's the same agent-era logic hitting your dinner order: less tapping, more telling the app what you mean and letting it do the rest.
Beat writer's block in your browser
Via ChordGen
ChordGen is a free, AI-powered chord progression generator that turns plain-language prompts like "melancholic jazz" or "upbeat pop anthem" into multiple genre-aware progressions. Hear them on an interactive piano, edit visually, and export to MIDI—no sign-up required.
How you can use it:
Turn a mood or genre prompt into ready-to-play progressions
Audition ideas on an interactive piano before committing
Explore unfamiliar styles without deep music theory
Export straight to MIDI for any DAW
Pricing: Free
An AI co-producer inside Ableton
Via VIXSOUND
VIXSOUND lives inside Ableton Live on macOS and responds to natural-language or voice chat to generate editable MIDI—melodies, chords, bass, drums—load instruments, separate stems, and handle mixing and arrangement. Because analysis runs locally and it outputs MIDI rather than finished audio, your files stay private, editable, and fully yours.
How you can use it:
Generate editable MIDI parts by voice or text, right in Ableton
Separate stems and transcribe audio to MIDI locally
Analyze tempo, key, and structure on the fly
Speed up production while keeping full creative control
Pricing: Paid
Jobs, announcements, and big ideas
ElevenLabs introduces AI avatars in ElevenCreative, turning text prompts into talking-head videos without cameras or actors.
Perplexity brings Deep Research into Computer, generating full reports and dashboards from a single query.
Meta adds an AI assistant and a desktop version to its Edits video app for creators.
OpenAI acquires Ona to expand Codex cloud environments for its developer coding tools.
Anthropic launches the Claude Corps Fellowship to bring AI benefits to underserved US communities.
One small step for man…one giant leap for Claude. Here’s what else to know in AI and future tech this week ⏯
That’s a wrap! See you next week for more.
