I was just talking about semantic_text and using the docs to show what I was talking about and have some feedback. The specific thing I was trying to talk through was the fact that the semantic_text field stores an inference ID in the mappings.

I started with the mapping reference page, which doesn’t have a single example specifying an inference endpoint. It does mention it’s using the default, and sends users to the tutorial. The examples still don’t show creating an inference endpoint, you have to read the prose in the document to see that that’s even an option. You have to go back to the mapping page and hunt for a link to parameters to see an example of how this works.

I know that we’ve done a lot of work to simplify these docs and make them more accessible, but I worry that we’re going too far, and the information that users actually need isn’t that accessible. I was able to pull the right information using Chat GPT, but that was not 100% correct (for example it said you couldn’t customize token pruning, which you now can in the mappings).

Maybe we can make this accessible as a quick resource w.r.t. what's available - one thing we could think about adding is a “fully functional” example with all the configuration to the top level page, and link to the params for explanations why? That would be one additional example, so not too onerous, but gives a quick high level “this is what I’m looking for” overview?

Stakeholders: @leemthompo @Mikep86