Is there a good reason that GPT3 DaVinci is the default LLM in the examples in the docs, not GPT-3.5?

[mlukewizard]

Title mostly describes it. Why is GPT3 DaVinci the default in the examples in the docs when ChatGPT is the newer model, much cheaper to run for testing out the library, and has better performance on almost everything.

Have we observed better performance with chains, agents etc with GPT3? Or is it just that we haven't got through updating the entire library and docs after the ChatGPT API was released?

This was helpful, #1376 but doesn't really answer the question about whether it's the intention to move over, or whether there's a good reason for staying with GPT3.