Thread by @rowancheung

OpenAI released their own Prompt Engineering Guide.

The guide is useful for anyone trying to maximize LLMs. I'd recommend it.

Here's the 6 strategies they outline for getting better results from GPT-4:

1. Write clear instructions.

Example: Specify the desired length of the output.

Prompt:
Summarize the text delimited by triple quotes in about 50 words.

"insert text here"

2. Provide reference text

Example: Instruct the model to answer using reference text.

Prompt:
<insert articles, each delimited by triple quotes>

Question: <insert question here>

3. Split complex tasks into simpler subtasks

Example: Summarize long documents piecewise and construct a full summary recursively.

I have very long, advanced prompt that helps anyone achieve this. I'm giving it away in my Advanced ChatGPT guide to new subscribers of my newsletter right now: therundown.ai/subscribe

4. Give the model time to "think"

Example: Instruct the model to work out its own solution before rushing to a conclusion.

5. Use external tools

Example: Use code execution to perform more accurate calculations or call external APIs.

6. Test changes systematically

Example: Evaluate model outputs with reference to gold-standard answers.

If you liked OpenAI's prompt guide, you'll probably find the Advanced ChatGPT guide useful.

-1000+ best ChatGPT prompts
-100+ AI tools
-25+ AI tool workflows

I'm auto-sending it free to new sign ups for my AI newsletter right now: therundown.ai/subscribe

Here's the link to the full prompt guide by OpenAI to read more:

Follow me @rowancheung for more AI tutorials.

If you found this useful, feel spare me a like/retweet/comment to support my content 👇 platform.openai.com/docs/guides/pr…

https://x.com/rowancheung/status/1736424525825196062?s=20