My Pickaxe asks Claude to review product sales pages and then create sales emails based on the product. He sometimes refuses to cooperate citing ethical concerns, usually because he says the product is explicitly adult, or is intended to mislead. In the cases I have witnessed, neither allegation is true. But I guess it may be that the cache he is using has issues because the pages have changed? I’ve told him to ignore any historical data and just use the current version of the webpage, but that doesn’t work. Is there a way round this?
Hi @steve_roberts I encountered this issue using the non-API version of Claud before subscribing to their pro plan. However, when I started using their API most of these issues disappeared. With that said, in my personal experience, most Pickaxes work better with GPT-4o.
Try 4o and test if the issues persist.
Hope that helps!
This sounds like a case where the context really matters! Generally speaking, you can prompt engineer around any issue. But you need to test it a lot and get a sense of when and why it’s not cooperating and then call those out in the prompt.
As ned_rvth suggests as well, using other models might circumvent the issue with less effort.
1 Like
Thanks for the input. I’ve realised that when you ask the API to interrogate a URL, it doesn’t actually do that: it looks at a cache. So on top of Claude not cooperating, there are some URLs that it just guesses about. I tried all the models - none of them can look at live websites and analyse them.