Open dymurray opened 2 months ago
I think this is a really good idea, probably a summary model, that can take websites and summarize them, then we add the results as AIMessages
to the 1Human1 or 1System1 Messages in the resulting agent calls.
Something like this: https://python.langchain.com/v0.2/docs/how_to/query_few_shot/ IIUC
When a rule references an external link, there is usually a ton of useful data and context for the LLM to scrape to provide better results. If we find an incident includes a link, consider scraping that data and tweaking our result based on it.