digest
Short Summary
Main recommendation: Use a four-part prompt formula, context plus specific information plus your intent plus the response format you want, to get the best result from ChatGPT, Bard, or Bing in one attempt instead of many revisions. The speaker says this formula came from seven prompt-engineering best practices and works across all three platforms.
Key Points
- The formula is: context + specific information + your intent + response format = the perfect prompt.
- Being specific matters most: a narrow question such as "What are the different breeds of small dogs suitable for apartment living?" gets more relevant answers than "Tell me about all the different dog breeds that exist."
- State your intent explicitly, for example "I'm helping my son with his science homework," so the output is tailored to that purpose.
- Specify the output format (a numbered list, step-by-step guide, or single paragraph) so the model knows how much and what shape of text to return.
- Correct spelling and grammar, since small errors can push the model in the wrong direction; the speaker suggests right-click spell-checking.
- Follow-up questions, rephrasing when needed, and a fact-checking prompt (feeding the output back and asking it to fact-check) are additional best practices the speaker uses alongside the formula.
Details Worth Noting
- Concrete instructions:
- Build each prompt from four parts in this order: (1) context, such as "I'm a beginner cook"; (2) specific information, such as "trying to make Italian cuisine"; (3) intent, such as "Can you provide me a simple and easy-to-follow recipe?"; (4) response format, such as "put the steps in numbered order from one through ten."
- Combined example: "I'm a beginner cook trying to make Italian cuisine. Could you provide me a simple and easy-to-follow recipe and put the steps in numbered order from one through ten?"
- Second example: "I'm a software developer working on a Python project. Can you explain how to implement exception handling in Python? Give it to me in a simple paragraph."
- Before the prompt, prime the AI; the speaker says to do this before asking any question and plans a separate video on prompt priming.
- If the result is off, experiment with different phrasing and reorder the pieces of the formula.
- The speaker claims the formula can save five to ten times the usual back-and-forth, though this is his estimate, not measured data.
- He reports studying prompting three to four hours a day since GPT-3 launched and taking courses; this is his stated background.
Bottom Line
Structuring each prompt with context, specifics, intent, and output format is the main lever the speaker offers for getting usable answers faster, and he presents it as platform-independent.
