APEPPERONI @dumbcluster

Markets as Mirrors: How Prediction Markets Become a Public Probability Layer I keep thinking about Luis Borges’ one-to-one map, which describes an ancient empire's futile attempt to create a perfectly detailed map that is the exact same size as the territory it represents. The…

AI “safety” isn’t safety - it's soft censorship. A few unelected execs shouldn’t decide what billions can build or see. DeAI flips the power: governance by users, not corporations. Guardrails without gatekeepers.

70% of your AI budget is really just a tax to centralized compute. Compute should be priced like liquidity on open networks: swap in, pay per second, swap out. Experiment faster. Win more.

80% of “vision” meetings are story time. Start with why. Assign who, what, and by when. Plans without dates are wishes. Dates without owners are lies. A due date is a promise. If no one owns it, you've scheduled a wish, not a commitment.

A mental model from the best founders: frugality drives invention. My takeaways from reading Bezos' ‘97 letter: - Customer obsession - Ruthless efficiency - Long-term vision - Frugality At @ionet: - Prioritize latency - Default automation - Activate idle GPUs in ~90 seconds.

Conviction plays are bets on shifting architecture. You have to see what no one else does and pick contrarian runners. I called internet, then mobile. Now, it's DeAI. Clouds can’t serve a trillion agents that hit ≈100k actions per user daily. Permissionless rails can.

Agents will eat UX. Design rails for machine-to-machine interaction: intent schemas, safety guardrails, and verifiable receipts. Frontrun the trend by building for invisible users. Which ones? AI agents, bots, scripts, and devices.

Work-life balance isn’t balance. It’s a trade. Early game, I front-load. I took exactly zero days off building @ionet. Not masochism. Momentum. The right to rebalance is earned later. And even then, I wouldn't pick it anyway.

Everyone frames AI as a race for models. But the bottleneck isn’t algorithms—it’s kilowatt-hours. Centralized clouds burn watts like jet fuel. Decentralized networks light up idle watts that have already been paid for. Winners won’t build bigger; they’ll use energy smarter.

The Kobeissi Letter @KobeissiLetter

5 days ago

Everyone frames AI as a race for models. But the bottleneck isn’t algorithms—it’s kilowatt-hours. Centralized clouds burn watts like jet fuel. Decentralized networks light up idle watts that have already been paid for. Winners won’t build bigger; they’ll use energy smarter.

424 1K 7K 738K 1K

Download Image

Beyond permissions, I'm worried more about limitations. One example is how AWS won't ever let anyone spin up 1000 GPT-5 instances. The reason? Too much compute in one pair of hands + they require you submit a custom form where they will ask for justification...

Tory | io.net 🦾 @MTorygreen

4 days ago

Beyond permissions, I'm worried more about limitations. One example is how AWS won't ever let anyone spin up 1000 GPT-5 instances. The reason? Too much compute in one pair of hands + they require you submit a custom form where they will ask for justification...

25 16 134 14K 9

Most of what we call “rigor” in scientific publishing is just waiting. Peer review isn’t quality control; it’s latency. Gated journals slow cures. Open preprints, reusable code, public datasets, and permissionless compute turn months into days and save lives. Lesson in there.

There’s one switch that turns AI from an app into an economy... what's the catch? Access. Centralized clouds gatekeep and price out builders. Go permissionless compute + open weights. You get cheaper runs, global reach, and intelligence that cannot be shut off.

80% of your “hustle” is untracked chaos. You 10x your outcomes with process, not luck. > Learn world-class systems. The internet gives you the keys. > Apply them ruthlessly. > Print the plan, underline the numbers, check them twice. Excellence is a habit you can audit.

Web3 doesn’t just need capital, it needs trust. Reputation is how we measure who shows up, who contributes, who builds. Without it, networks get lost in bots and noise. With it, real people lead the way.

In 10 years, Netflix will be gone. You’ll wake up and an AI will make a show just for you—based on your moods, likes, and dreams. Sound wild? Watch this:

Failure isn’t a scar: it’s a lever. I failed an assessment run in West Point by 11 seconds after an injury and didn't make the cut. The lesson wasn’t to just get up. It’s to get up faster. Your future is defined by how quickly you reset.

Originality isn’t a strategy; it’s a delay tactic. Elite execution is simple: 1. Watch the best in the world 2. Run their playbook 3. Test 4. Keep what sticks 5. Trim the rest I use standups, OKRs, and 1:1s because the numbers say they work.

Decentralization isn’t ideology. It's logistics, plain and simple. Centralized vendors slow decisions and tax runways across shipping, media, software, manufacturing, and even power distribution. Shift to permissionless tools. Ship faster. Burn less.

Bitcoiners think banks are evil. Not even close. Whatever power you think centralized banks have, I promise you centralized AI will be 100x worse. They will literally control your mind. That’s the real fight here.

Everyone thinks crypto is just about money. But money was only the start. The real purpose is much bigger: to decentralize intelligence. DeAI isn’t just an app—it’s why crypto was built.