Good morning. Most people think death means finally escaping social media. Not anymore. Meta has patented tech that lets AI run your accounts long after you’re gone. The bots will like your friend’s vacation pics, respond to comments, send DMs and even make video calls from beyond the grave. Meta’s not rolling this out anytime soon, but even floating the idea raises big questions about digital immortality. Soon everyone will be faking death to automate their socials.
AI that continuously learns isn’t real (and VCs don’t care). Investors keep opening their wallet for startups claiming they have an AI that learns like people do. But what they’re really delivering are models that use scratchpads, ‘memory pools’ that are just databases, and clever caching tricks. None of these methods teach AI anything new, and they can’t separate facts from hallucinations. Continuous tuning exists, but someone still has to manually collect and organize all that data. True continual learning would require rebuilding AI from scratch, and nobody’s even close. As long as VCs keep signing checks, reality can wait. [The Information]
Google just dropped an AI music generator. With the Gemini app and Lyria 3, anyone over 18 can spin up 30-second songs from text, images, or videos. Describe your mood, and Gemini makes a track to fit the vibe. Want a R&B slow jam about socks finding love in the dryer? Done. A disco funk track about tacos and pizza? Easy. And just when you thought human creativity peaked, Nano Banana throws in AI cover art. Sit back and watch those royalties roll in. [Google]
Richard Socher and Bryan McCann are among the most-cited Al researchers in the world. They just released 35 predictions for 2026, including:
- Capital floods back to research
- “Reward engineering” becomes an in-demand job
- Traditional coding will be gone by December
Retail investors can now bet their savings on SpaceX and Anthropic. Previously, investing in private tech startups was reserved for wealthy insiders and venture firms. Not anymore. Robinhood and Powerlaw promise to democratize Silicon Valley by letting regular people buy shares directly from company insiders. Great in theory, only these funds have tricky structures, delayed profits, and companies that keep financial details secret. A similar fund recently went bankrupt, too. But hey, at least you can brag about owning SpaceX. [The Information, Bloomberg]
AI is getting scary good at hacking smart contracts. Over $100 billion in crypto assets rely entirely on these contracts. Paradigm and OpenAI created a new benchmark called EVMbench to test how well AI detects, patches, and exploits major vulnerabilities. OpenAI’s GPT-5.3-Codex hit a mind blowing 72.2% success rate, more than 2x what models could achieve just months ago. Let that sink in. This directly impacts prediction markets, DeFi, DAOs, stablecoins, and pretty much all of Web3. AI agents might someday guard your crypto wallet, but right now they’re much better at cleaning them out. [OpenAI]
Democracy is drowning in AI-generated activism. Southern California’s air quality board recently rejected rules to phase out polluting gas appliances. Why? They got flooded by over 20,000 emails generated by the AI platform CiviClick. When checked, some were from people who hadn’t written in and had no idea their names were being used. Normally, the agency receives under 10 comments per issue, making this a tsunami of fake outrage. Experts warn it’s just the start of digital astroturfing, where AI blurs the line between real public opinion and automated spam. [LATimes]
fun stats
💰 $100 billion+. Record funding round OpenAI is finalizing with Amazon, SoftBank, Nvidia, and Microsoft, pushing its total valuation toward $850 billion. ChatGPT isn’t cheap.
🏰 $4 billion. Target valuation for Ineffable Intelligence, a new London AI lab led by ex-Google DeepMind scientist David Silver. Its $1B raise would be Europe’s largest seed ever.
👍 $65 million. What Meta is spending this year to back AI friendly state politicians and keep states from regulating data centers.