We should create a documentation page about best practices for working with the model.
Tip 1: Remind the model about the question/problem being asked
- For long notebooks, its easy for earlier parts of the document to not be included in the model prompt
- This often happens if you have very verbose output (e.g. you dump logs, or a really large JSON file)
- One outcome of this is if your are in the midst of troubleshooting some problem or trying to answer some question, the question might not be included in the input
- In this case you can remind the LLM about the question
- e.g. add a markdown cell and do something like "Based on the output above, does the GCP service account have permission..."
We should create a documentation page about best practices for working with the model.
Tip 1: Remind the model about the question/problem being asked