https://www.deeplearning.ai/short-courses/chatgpt-prompt-engineering-for-developers/
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/1/introduction
- Fine tuned instructions
- RLHF: Reinforcement Learning with Human Feedback
- More widely used than Base LLMs
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/2/guidelines
- Use delimiters
- Ask for structured output
- Provide concrete steps for the model to follow
https://twitter.com/SeanPlusPlus/status/1653842628067328000
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/3/iterative
- Build up prompts to get to a desired output
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/4/summarizing
- If you have a site with many reviews, super useful to ask LLM to summarize
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/5/inferring
- Useful tools for sentiment analysis
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/6/transforming
- Translating languages super straightforward
- Can also translate tone (slang to formal business)
- Format (JSON to HTML)
- Spellcheck and Grammar check
- Just ask it to use APA as well
- Can redline text and as well!
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/7/expanding
- Degree of randomness of model
- Temperature allows us to change variety of responses
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/8/chatbot
- Each interaction with a large language model is a standalon interaction
- You need to pass the entire body of previous parts of a conversation to "simulate" that a conversation is ongoing
- Also can program roles
- System, Assistant, User ... 3 roles
- As a dev, we can set the
System
message to "whisper in the ear" of theAssistant