I’m trying to create a tool that can pull live data and cite it.
The data that comes back looks good, but when you click through, the page doesn’t exist, or if it does, the data that is cited in the pickaxe output doesn’t actually exist on the URL.
I’ve rebuilt the tool a couple of times and spent 6 or 7 sessions trying to update the original 3 questions to ensure that the errors won’t occur.
No matter what I’ve tried or how I’ve updated the rules and output pieces, nothing seems to work. I’ve also tried different LLM’s without success.
Any suggestions or insights welcome!
The easiest way to avoid hallucinations would be to use an action to actually pull live data as opposed to what sounds like the query you are running instead.
If you share more about what you are trying to do and how you are doing it, we might be able to be of greater assistance.
It appears you are experiencing problems with citation accuracy, perhaps because there is dynamic content on the pages or problems with the URLs that are being generated. One thing to look for is whether the URLs you’re extracting are in correct form and whether the content is being rendered dynamically (e.g., via JavaScript) that perhaps the tool you are using is not processing. If you’re pulling or scraping data, it may be worthwhile to use a headless browser (such as Selenium or Puppeteer) that can execute pages rendered by JavaScript. Also, ensure that your tool is not pulling stale or cached data and is pulling live content every time. Debugging the relationship between the page and the data may determine if the problem is in the data or the page.