2026.08.06 // LLM FOR DUMMIES // 7 MIN
We have the machine. Now, why are its trillion numbers right? Because of one absurdly simple game played ten trillion times: hide the next token, guess it, measure the surprise, adjust. This part explains why that game forces a model to learn facts, grammar and a bit of reasoning — and what a loss curve really measures.
→
2026.07.30 // LLM FOR DUMMIES // 9 MIN
Parts 1 and 2 turned text into lonely points in space. Attention is the mechanism that lets every token look at every other token and decide whom to listen to — the idea the Transformer is named after, and the reason long context windows cost quadratic money.
→
2026.07.26 // LLM FOR DUMMIES // 8 MIN
Part 1 left us with a list of opaque numbers. Now the magic starts: every token becomes a point in a space with thousands of dimensions, where distance is similarity and directions are concepts. This is the idea powering semantic search, recommendations, RAG — and the input layer of every LLM.
→
2026.07.22 // LLM FOR DUMMIES // 9 MIN
First stop in a series that explains how large language models actually work, no PhD required. Before any 'intelligence' happens, your text gets chopped into numbered chunks called tokens — and half of the weird things LLMs do trace back to exactly this step.
→
2026.07.22 // AI & REGULATION // 3 MIN
In eleven days the AI Act's high-risk obligations become enforceable: conformity assessments, CE marking, €35M fines. Meanwhile 78% of enterprises haven't taken meaningful compliance steps and a proposed delay to 2027 is still not law. The gap, in three charts.
→
2026.07.20 // MARKETS // 3 MIN
The SOX fell 24% in three weeks. Intel lost a fifth of its value, Korea hit circuit breakers, and Samsung dropped 7% on a +1,800% earnings print. What the semiconductor correction does—and doesn't—tell you about the AI trade.
→
2026.07.19 // AI & PAYMENTS // 3 MIN
AI agents already influence $67B in retail sales and 20% of orders—then usage collapses from 62% at product comparison to 23% at checkout. The $3-5 trillion agentic commerce market is blocked on payments, and that gap is the story.
→
2026.07.18 // AI & MARKETS // 4 MIN
US data-center power demand is set to double by 2027, PJM capacity prices are up 11x in two years, and your electricity bill is co-financing the AI boom. The bottleneck moved from chips to megawatts—in three charts.
→
2026.07.17 // DIGITAL ASSETS // 4 MIN
A $322B market cap, on-chain volumes rivaling Visa, and tokenized Treasuries at their 10th consecutive monthly record. The most underpriced infrastructure story of 2026—in three charts.
→
2026.07.17 // AI & MARKETS // 4 MIN
Four companies will spend more on AI infrastructure in 2026 than most countries spend on everything. The money is real. The revenue is projected. The deals are circular. Three charts on the most crowded trade on Earth.
→
2026.01.02 // AI & TRADING // 4 MIN
Only 11% have agentic AI in production. Here's what separates winners from the 89% burning cash on experiments—with data from McKinsey, Deloitte, and lessons from early failures.
→
2025.02.17 // TECHNOLOGY // 2 MIN
Artificial Intelligence is no longer just a buzzword in finance. It is increasingly becoming the engine that powers modern financial markets.
→
2025.02.17 // FINANCE // 2 MIN
The new rule introduces a tax revaluation of crypto assets held as of January 1, 2025, through the payment of a reduced substitute tax.
→