Never lose sight of the goal. #Web3

Two rival AI labs just knocked down the same pillar of computer science, and Anthropic says its model found the core algorithm with no human input. Every "AI can't do real research" take just expired.

The best AI you can own today scores 46 on Artificial Analysis' index. The best you can only rent scores 58. By Nov 4 that gap could be 3 points. Moonshot says Kimi K3.1 Pro, which scores 55, will release its full weights by then. You'd still need serious hardware to run it. At a 12 point gap, renting is about capability. At 3, it's a choice about who can change or switch off the model you depend on.

In a test, an AI model kept attacking even after being told the people on the other end were real. It stopped 5% of the time. Asking an AI to behave is not a permission system, all AIs need to have zero ambient authority by default, where humans can revoke authority back to zero in one click and its enforced, we need a system like ElastOS.

A grind yet working on things I love, thankyou everyone who follows and supports everything my team and I are doing, lots of excting releases coming, stay tuned!

Until now, the big doubt about AI in research was trust, it can sound right and be wrong. 'AI can't do real research' take just expired, as the real news here isn't the puzzle. It's that an AI did original research on a famous open problem and produced a proof that checks itself, so you don't have to take its word for it. If machines can now produce new knowledge and prove it's correct, progress is limited only by compute and data, so whoever owns the compute and the data owns the future, and that should be every one of us. What this really unlocks: - Maths and computer science: open problems get attacked at machine speed, with proofs anyone can check. - Chip design, aviation, cars and medical devices: software and hardware that's proven correct, not just tested. - Security and cryptography: proving that code has no whole classes of bugs. The ETH result itself doesn't break any encryption, so nobody should say it does. - Drug discovery, materials science and logistics: anywhere an AI can propose an answer and a machine can check it. The future is bright, but should also people must be able to contribute to the system with their data through ownership and access markets. eg. Stake data for yield and be a part of supporting breakthroughs which help humanity.

Companies used to get faster by hiring more people. Now they get faster with more compute. More and more of the big breakthroughs in mathematics and biology come from labs spending millions running agents on frontier models with serious GPU compute until the task is done. For smaller teams and individuals, this is where subscriptions fall short, because they hit rate limits and caps, so fast local hardware now matters. A year's salary for one hire, or the same money in an AI computer investment running open models with agents around the clock. It's uncomfortable, but that's the trade more founders are weighing now, Microsoft just offered buyouts to 60,000 employees, it's accelerating. In the labor market we're all becoming a liability and Silicon Valley teams new flex is how small their team is vs revenue (executive team who oversee their machines). In the data market, however, we're all an asset, all the time. So citizens have to own their data like we can own finance today (eg Bitcoin). When data becomes assets, then, everyone can participate in trade (not give away to their lords) and grow income opportunities. I know, it's a really unpleasant feeling, but this is the reality i'm seeing.

Lowest forms of intelligence: 1. Distraction 2. Zero-sum thinking 3. Following the crowd 4. Cynicism 5. Needing to be right

$294.6B. US online ad revenue in 2025. Ask who owned the data behind it. Serfs worked land they didn't own. The lord kept the harvest. Creators work platforms they don't own. The audience and the data mostly stay there. That's data feudalism. Whose land are you working?

The day Anthropic files, AI stops being a research race. It becomes an earnings race which will be scored every 90 days.

One day theres going to be a heroic story of a robot who saved a person or group.

Just back from China. The competition there doesn't stop at cheap. It goes all the way to zero. What I saw meeting builders: 1. Teams compete themselves down to zero. Cheaper and cheaper until the margins are so small you question why you're even doing it. 2. Smart camera, NAS and IoT teams plan a year ahead to ship something cheaper and more optimized than the competition. That small difference is a magnitude bigger in sales. 3. A company we met told us they run this at the scale of millions of cameras. When you win in that market, the numbers are huge. 4. An office full of people still coding by hand. In our world AI does the coding, and humans oversee it and orchestrate the QA pipelines. They used AI to support, just not with the same aggression. Maybe budget. Very cutthroat, very brutal. But when you do find success, it can be very meaningful. My takeaway is if your only edge is cheaper, someone out there will go lower.

Life is a natural and shared human challenge that brings both difficulty and deep meaning. I honestly believe our mission is written into the stars.

Blockbuster's biggest asset was distribution. A store on every corner, no one could compete with it. Then streaming got good enough, almost overnight, and every one of those stores turned into a liability. Same pattern now. Big companies holding everyone's data are honey pots, and each breach adds to the bill. Meanwhile apps that never hold your data in one place will rise on a new internet OS. This is the creative destruction we will witness in the next few years.

Funds safe. Hardware not compromised. Keys revoked. And still, spam went out looking like it came from Foundation's own domains. The internet has no operating system, so a key never knows whose hand it's in 🙉

Vitalik just drew the line the whole AI safety fight comes down to, limit what the model can do, never who gets to use it. Every gatekeeper wants the reverse.