Why is my Pickaxe generating content from non-existent webpages?

I am using openAI Chat GPT4o to generate content and also using GPT4o for my created pickaxe named “Product Description Generator.” I am using both to compare the quality of the output and I’m getting completely different and useless outputs from PickAxe using the same source URL.

I am requesting the openAI chat extract the trademarked, registered trademarked, and other features of a product. It finds them easily and generates an excellent product description.

Pickaxe is given the same information to query and it generates product features that don’t even exist on the page. It doesn’t extract, or identify any of the trademarked, registered trademarked, and other features of a product.

I have made adjustments to the role, frame, and settings in an attempt to obtain similar output as openAI. Pickaxe keeps generating content that does not exist on the page.

Please advise! Thanks!

1 Like

this is expected. there’s stuff going on in the background of pickaxe in the API. settings will play a big role. Are you using a knowledge base at all?

Hi @visionintodestiny this is a common issue but the cause varies depending on the application and several other factors.

If you can send me some screenshots/screen recordings I can take a look for you. Feel free to send me a DM. Thx

Pickaxe will not be able to access a webpage unless you enable web-browsing. You can do this by connecting the Google Search action. You can find this in the Act tab.

ChatGPT is OpenAI’s own implementation of their model. It is basically a super general-purpose Pickaxe that they built on top of their own LLM. Then they give it a few custom actions (web browsing, code interpretter, image generation). Their base LLM cannot do those things by itself. It needs the connected tools. On Pickaxe you’re using the base LLMs and then can connect your own actions.

1 Like