hawkes.aiKey ideas
- The one idea — The contract: no unexplained words, one idea per episode, failures on the syllabus, simplifications flagged — in exchange for watching in order.
- How it is shown — The contract itself, clause by clause.
- The trap to avoid — Courses that only show wins are marketing.
- What it sets up — A safety experiment of ours that failed.
Any AI course that only shows you the wins is marketing. This one puts failures, costs, and security holes on the syllabus — and states its own limits.
The one idea
The contract: no unexplained words, one idea per episode, failures on the syllabus, simplifications flagged — in exchange for watching in order.
Before you give this series eleven hours of your life — here's the contract. Two minutes of rules, then eleven hours of machine. Rule one: no unexplained words. Every term is built from parts you already have. When "logit" arrives, it will be because an episode earned it. Jargon from nowhere is a lie about how understanding works. Rule two: one idea per episode. Three minutes, one mechanism, one concrete demo, one trap people actually fall into.
How it works — the demo
The contract itself, clause by clause.
If an idea needs two episodes, it gets two. Rule three: the ugly parts are on the syllabus. Limits, costs, failure modes, security holes — and near the very end, a safety experiment of our own that failed. A course that only shows wins is marketing. You'll know this one's limits because it will state them. Rule four: when we simplify, we say so — and we come back to fix it. Metaphor on the first pass, mechanism on the second. Your end of the deal: watch in order.
The trap to avoid
Courses that only show wins are marketing.
Why it matters — and what’s next
A safety experiment of ours that failed.
Every episode stands on the ones before it. Hold that, and the payoff is real: from "I used a chatbot once" to explaining how frontier models are trained, served, constrained, and audited. Not vibes — the machine. Skip around and it breaks: episode ninety assumes twelve; episode two hundred assumes ninety. That's the whole promise. Signed. The climb starts now — at the bottom of everything: a machine that cannot even see letters.
What would it take to go from 'I used a chatbot once' to explaining how frontier models are trained and served? Four rules, kept for eleven hours.
This is one short episode in AI: Zero → Frontier, a step-by-step climb through how AI actually works. Each episode builds only on the ones before it.
Full transcript 2:00 of narration
Before you give this series eleven hours of your life — here's the contract. Two minutes of rules, then eleven hours of machine.
Rule one: no unexplained words. Every term is built from parts you already have. When "logit" arrives, it will be because an episode earned it. Jargon from nowhere is a lie about how understanding works.
Rule two: one idea per episode. Three minutes, one mechanism, one concrete demo, one trap people actually fall into. If an idea needs two episodes, it gets two.
Rule three: the ugly parts are on the syllabus. Limits, costs, failure modes, security holes — and near the very end, a safety experiment of our own that failed. A course that only shows wins is marketing. You'll know this one's limits because it will state them.
Rule four: when we simplify, we say so — and we come back to fix it. Metaphor on the first pass, mechanism on the second.
Your end of the deal: watch in order. Every episode stands on the ones before it. Hold that, and the payoff is real: from "I used a chatbot once" to explaining how frontier models are trained, served, constrained, and audited. Not vibes — the machine. Skip around and it breaks: episode ninety assumes twelve; episode two hundred assumes ninety.
That's the whole promise. Signed. The climb starts now — at the bottom of everything: a machine that cannot even see letters.