mistercrunch commented on PR #32649: URL: https://github.com/apache/superset/pull/32649#issuecomment-2790792306
I think this is an interesting proof of concept, but to me it raises more questions than it provides answers. Raising some of the questions here: - seems we're deciding to tightly couple with langchain/openai here, bringing new dependencies to the core project for something that's being a feature flag, is that the path forward? what about other LLMs? do we want to bundle langchain and commit to it as the bridge to LLMs? are there other toolkits we should consider before committing to langchain? - the design is a bit rough (injecting a prompt over the chart title), potentially could be many more LLM "touchpoints" in the app, what's the pattern to expose AI features in the UX? What's the better pattern to trigger and expose this particular feature? - sharing data/metadata with external providers: I can see how different environments may or may not be cool with sharing either metadata and/or data here. FF would have to come with heavy disclaimers. "THIS FEATURE FLAG WILL ALLOW USERS TO SHARE SENSITIVE DATA WITH THE PROVIDER ... " - the prompt is good, but there's no such thing as a "perfect prompt", so should the prompt be configurable too? maybe that'd need to be exposed as a configuration parameter as well? - wider context: do we need an env level prompt preamble "This instance of Superset is set up for CompanyX, we care a lot about ..." - no vector db: here we're passing some of the current context for these types of features in the pass we might need a vector db - do we need to know whether/how this fits in as a foundational part of the wider "LLM interaction framework" as it evolves? My intuition around much potential AI features - given that there are a hundred ways to implement them depending on prefs / environment constraints - and that design patterns (both UX and backend-wise) are still evolving rapidly - is that they probably belong in extensions, at least for the time being. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
