It'd be even better if it can provide links to its sources.
However looking at the code I suspect I know why it doesn't: it's using just the search results with the snippets as its source, and the LLM is doing a lot of heavy lifting generating the content, meaning it's mostly synthesised and more likely to be wrong (particularly with temp=0.7)
So a nice increment on this would be to use a scraper like firecrawl to extract the page content and use that for the report instead of the search results.
Also this would have the benefit of being able to include references.
Hi this is really great prompt engineering!
It'd be even better if it can provide links to its sources.
However looking at the code I suspect I know why it doesn't: it's using just the search results with the snippets as its source, and the LLM is doing a lot of heavy lifting generating the content, meaning it's mostly synthesised and more likely to be wrong (particularly with temp=0.7)
So a nice increment on this would be to use a scraper like firecrawl to extract the page content and use that for the report instead of the search results.
Also this would have the benefit of being able to include references.