The “leaked” Anthropic prompting video. It's free.
A post going round claims a $1.2m engineer smuggled out Anthropic's prompting playbook. It's their own class, on their own channel. Here it is, with the 5 moves worth stealing.
By Abi Odedeyi · CodeFreeIQ
🎬
The actual video
Prompting 101, taught by the Applied AI team who do this for a living.
📝
My 5 running notes
Watch nothing twice. The five moves that matter for a business owner.
🆓
No leak, no course fee
Free because it always was.
Get the video + the notes
Pop your details in and it unlocks right here.
No spam. Unsubscribe anytime.
The video
Anthropic's own Prompting 101, taught by Hannah and Christian from their Applied AI team. Never leaked, always free.
My 5 notes, so you don't need a notepad
1. Order matters. Role and task first, then your documents, then step-by-step instructions, then examples, then repeat the one thing that matters most, then "go". That structure is the whole class.
2. The stuff that never changes goes in the system prompt. Your service menu, your form, your rules. Claude stops re-working it out every single time.
3. Label your sections. Wrap each block in a named tag like <pricing> or <transcript> so it knows what it's looking at and can refer back.
4. Give it an out. Their exact anti-hallucination move: "only answer when you're confident, and say when you can't tell." Without it, it guesses politely.
5. Tell it what order to read things. In their demo it reads the form before the sketch, like a human would. And when it gets a hard case wrong, that case goes back in as an example.
Bonus: start the answer for it ("begin your reply with <verdict>") when you need a strict format.