AI & Technology

Sam Altman says AI will transform education. The venture funding data tells a different story — the money goes to habit-forming practice apps, not tools that verify real learning. Meanwhile the US, EU, and China are running three genuinely different bets on how AI belongs in the classroom, and none of them are finished. This is a status check on where things actually stand, not another prediction piece.

Building AI sounds like something only a handful of billion-dollar labs can do. It isn't — using AI, building with it, fine-tuning it, and even training a narrow model from scratch are all realistic for one motivated person. This piece breaks down the five real levels of "building AI," with real companies (Cursor, Harvey, Sanofi, Microsoft) at each one, plus a step-by-step guide to building your own AI-powered tool this week.

While Mars gets the headlines, four governments are quietly racing to claim and defend a layer of infrastructure a few hundred kilometers overhead — and none of them are doing it entirely alone.

Roughly 97.5% of drug candidates die before a patent is ever filed on them — invisible to competitors, regulators, and the public. This piece breaks down how AI is starting to mine that hidden graveyard, and why no single unpatented failure has turned into a headline win yet.

Warren Buffett says he could still compound money at 50% a year. The data says the market that let him do that in the 1950s doesn't exist anymore — and here's the specific evidence why.

Rare earth materials show up across the AI infrastructure stack — in chips, chipmaking equipment, and power/cooling systems — but exposure varies sharply by company. Here's a full-stack look at where dependency is highest (TSMC, Micron, ASML), where it's lowest (Vertiv, GE Vernova, Eaton), and how to think about diversifying an AI infrastructure allocation accordingly

Rare earths aren't actually rare — deposits exist on every continent. What's concentrated is processing, a capability built over decades. Here's how the mining-to-magnets supply chain works, where global diversification efforts stand, and how investors can gain exposure to both the established leader and the emerging alternatives.

AI in pharma is often described as one trend. It's really two, and it only wins one on its own merits. In drug discovery, AI directly attacks the problem of finding the right molecule, letting small teams compete with giants using nothing but data and talent. In drug manufacturing, AI can't shortcut anything, since it depends on years of proprietary production data that only exists once a company has already built an expensive facility, for reasons that have nothing to do wit

Angel checks range $25K-$350K, VC spans $500K seed to $200M late-stage, PE requires $50M+ ARR — and investment banks aren't investors at all. A full map of who funds what, at each stage, and how to actually reach them.

AI drug discovery has one validated success and zero FDA approvals so far — while AI in trial design already has formal regulatory qualification from both the EMA and FDA. The most-hyped application in pharma AI isn't the one furthest along.