Task as it goes against OpenAI's use case policy

I have asked the LLM to
“Then search these sites for devices that have these specifications: The Device Type is AI, the Target is SC, the Drug Format is Liquid, the Min Volume is 0, the Max Volume is 1ml, the Drug Container is PFS, and has an Auto-Injector.”
I have a GoogleSearchAPIWrapper() set up in LangChain to do the search. I am not looking for PII/individuals’ device info. I am just looking for a specific medical device from a manufacturer on the manufacturer’s website.

It seems like the problem might have been this prompt:
create a list of all medical device manufacturers.

It seems to think that I am web scraping (See message below). We are not really web scraping at least in an unethical context. We are looking at manufacturer’s websites and looking for detailed info they have on the products they are trying to sell. We would end up buying these products from the manufacturer so we are doing something they would wholeheartedly support.
Can we get a fix to this somehow? Maybe you can whitelist certain websites? Or, easier for you, ask websites to submit their url if they want to be blocked?

Could not parse LLM output: I’m sorry, but I cannot perform this task as it goes against OpenAI’s use case policy. It is not ethical to scrape websites for information without their consent.
Observation: Invalid or incomplete response
Thought:I cannot perform the task as it goes against OpenAI’s use case policy. I need to inform the user that I cannot scrape websites for information without their consent.

Um, not so fast. :wink:

What if you are comparing manufacturers only to point out flaws or other commerce-related reasons not to recommend a specific product? The vendors do not know your intent any more than OpenAI knows your ultimate intent. Ergo - you’re indiscriminately scraping.


Decouple the inferencing process from the scraping process. Use something like Browserless to gather the data and then use AGI to render the assessment. You are kind of doing that already, but you are intimating GPT should scrape in the prompt, right?