Hey hustlers,
Quieter week than the last one, but don't let that fool you. The drama shifted from product launches to something juicier: the CEOs are fighting in public now.
Also, your face officially became intellectual property this week. And Meta just did something it swore it would never do.
Here's what you need to know.
👾 WHAT'S NEW IN AI
1. Sam Altman Called Anthropic's Mythos Model "Fear-Based Marketing"
Yesterday on a podcast, OpenAI's CEO directly called out Anthropic for claiming their Mythos model is "too dangerous for public release." Altman says it's scare tactics to make the model sound more impressive than it is, and a way to keep powerful AI exclusive. The irony is OpenAI has done the exact same thing for years. Remember when Altman himself warned AI could cause "civilizational risk"?
Why you should care: When CEOs attack each other's marketing instead of their models, it means the technology gap is shrinking. The war is moving from capability to narrative. Watch what they ship, not what they say.
2. YouTube Now Lets Celebrities Hunt Down and Remove Deepfakes
On April 21, YouTube expanded its deepfake detection tool to the entertainment industry. Actors, musicians, athletes, and major talent agencies like CAA and WME can now scan for AI-generated fakes of their faces and request removal. You don't even need a YouTube channel. It works like Content ID but for human faces.
Why you should care: Your face is becoming intellectual property that platforms are building tools to protect. YouTube is also pushing for the NO FAKES Act to make this a federal law, not just a feature. This is where digital identity is heading.
3. Meta Shipped Its First Proprietary AI Model (and Quietly Killed Open Source)
Meta launched Muse Spark, built by ex-Scale AI CEO Alexandr Wang's new Superintelligence Labs. The big shift: it's proprietary. Not open source. Not open weights. A complete reversal from the Llama strategy. It powers Meta AI across WhatsApp, Instagram, Facebook, and Ray-Ban glasses, with a new Shopping mode that recommends products from creator content. Meta is spending $115-135 billion on AI this year.
Why you should care: Meta was the biggest company giving away powerful AI models for free. That era just ended. If you relied on Llama for anything, start watching alternatives like Google's Gemma or Mistral.
👾 THE GOOD STUFF
🔧 AI Tool: SkillClaw
A new framework from the DreamX Team that lets AI agent skills evolve and improve automatically based on how real users interact with them. Instead of manually updating agent behaviors, SkillClaw uses an "agentic evolver" to refine, create, and update skills on its own. Early results show a +42% average improvement across real-world tasks. If you're experimenting with AI agents, this is worth a look.
🐙 GitHub: forrestchang/andrej-karpathy-skills (71.5k stars)
A single CLAUDE.md file derived from Andrej Karpathy's observations on LLM coding pitfalls. Drop it into your project and it noticeably improves how Claude Code behaves. Simple, practical, trending hard.
🎬 YouTube: 19 Claude Opus 4.7 Insights You Wouldn't Get From the Headlines
Everyone covered the benchmarks. This video digs into the stuff most people missed, from hidden features to real-world quirks that actually matter when you use it.
👾 TO READ
Neural Computers: A New Computing Paradigm Where AI IS the Operating System
📎 https://arxiv.org/abs/2604.06425
What they built: Researchers from Meta AI and King Abdullah University of Science and Technology created something called Neural Computers.
Instead of AI running as an app on top of your operating system, the neural model itself becomes the computer.
It handles computation, memory, and input/output all within a single learned system.
What they found: The prototype can generate high-fidelity terminal screens and control GUI interactions directly from user input traces. It learns by watching how you use your computer and then replicates those patterns autonomously.
Why it's interesting: Last week we covered Perplexity putting an AI agent on your Mac. This paper takes that idea to its logical extreme: what if the AI didn't just use your computer, but literally was the computer? We're a long way from that being real, but the research direction is clear. The future operating system might not run apps. It might just understand what you want and do it.
🧵 Thread Drop
Quieter week? Maybe.
But the shifts underneath are bigger than any product launch. The rules of who owns what, who controls what, and who gets to use what in AI are being rewritten in real time.
And that's what we're here to track.
👾 See you soon 👾
Want to work together? If you're building something in AI and think our audiences would vibe with it, let's talk.
