hawkes.aiKey ideas
- The one idea — Five common mistakes — vague asks, no context, no examples, no role or goal, and treating one reply as final — all trace to a single misunderstanding: the model is a next-token predictor that only has your text, so you fix it by feeding it the material, not by hunting for magic words.
- How it is shown — The same machine fed two requests — a thin wisp of fog versus a rich, specific crystal — producing mush versus a clean sculpture; then the five fixes applied one at a time.
- The trap to avoid — Chasing "secret prompts" and magic phrases — engineering (giving it context, examples, and a clear target) beats any incantation.
The same AI, the same day — two people, wildly different results. One gets vague mush; the other, exactly what they needed.
The one idea
Five common mistakes — vague asks, no context, no examples, no role or goal, and treating one reply as final — all trace to a single misunderstanding: the model is a next-token predictor that only has your text, so you fix it by feeding it the material, not by hunting for magic words.
The difference isn't a secret model or a magic word. It's five habits. Here's how to stop using AI wrong. Mistake one: the vague ask. Remember what this is — a next-token predictor working only from the words you give it. It can't read your mind or your intent. Ask it to "make this better" and it has to guess what "better" means. Say what you actually want, and the guessing stops. Mistake two: no context. You know your situation — the audience, the history, the document you're working from. The model knows none of it unless it's in the window.
How it works — the demo
The same machine fed two requests — a thin wisp of fog versus a rich, specific crystal — producing mush versus a clean sculpture; then the five fixes applied one at a time.
It isn't lazy; it literally cannot see what you didn't paste. So paste it — the email, the draft, the constraints. Grounding beats guessing. Mistake three: describing instead of showing. You can spend a paragraph explaining the tone and format you want — or paste one example that nails it. Because it learns the pattern from what's in front of it, a single good example steers it harder than any description. Show it the shape you want. Mistake four: no role, no goal. The model will happily produce a competent, generic answer aimed at no one. Tell it who the reader is and what good looks like — "for a skeptical CFO, one page, no jargon" — and every next word bends toward that target. You're not flattering it.
The trap to avoid
Chasing "secret prompts" and magic phrases — engineering (giving it context, examples, and a clear target) beats any incantation.
Why it matters — and what’s next
You're aiming it. Mistake five: treating the first reply as final. It's a draft machine, not an oracle. The real power is the second turn, and the fifth — "too formal, cut it in half, keep the third point." And forget secret prompts: no magic phrase unlocks a better model. Engineering beats incantation. Five mistakes, one cure: stop treating it like a mind reader and start handing it the material — context, examples, a clear target — then refine. That's the whole skill. But there's a catch we tackle next: a well-fed model can still be confidently, fluently wrong. Knowing when it's guessing.
This is a supplement in AI: Zero → Frontier — a side-trip that deepens the act it sits beside, one file and one loop at a time.
Full transcript 2:45 of narration
The same AI, the same day — two people, wildly different results. One gets vague mush; the other, exactly what they needed. The difference isn't a secret model or a magic word.
It's five habits. Here's how to stop using AI wrong. Mistake one: the vague ask.
Remember what this is — a next-token predictor working only from the words you give it. It can't read your mind or your intent. Ask it to "make this better" and it has to guess what "better" means.
Say what you actually want, and the guessing stops. Mistake two: no context. You know your situation — the audience, the history, the document you're working from.
The model knows none of it unless it's in the window. It isn't lazy; it literally cannot see what you didn't paste. So paste it — the email, the draft, the constraints.
Grounding beats guessing. Mistake three: describing instead of showing. You can spend a paragraph explaining the tone and format you want — or paste one example that nails it.
Because it learns the pattern from what's in front of it, a single good example steers it harder than any description. Show it the shape you want. Mistake four: no role, no goal.
The model will happily produce a competent, generic answer aimed at no one. Tell it who the reader is and what good looks like — "for a skeptical CFO, one page, no jargon" — and every next word bends toward that target. You're not flattering it.
You're aiming it. Mistake five: treating the first reply as final. It's a draft machine, not an oracle.
The real power is the second turn, and the fifth — "too formal, cut it in half, keep the third point." And forget secret prompts: no magic phrase unlocks a better model. Engineering beats incantation. Five mistakes, one cure: stop treating it like a mind reader and start handing it the material — context, examples, a clear target — then refine.
That's the whole skill. But there's a catch we tackle next: a well-fed model can still be confidently, fluently wrong. Knowing when it's guessing.