AI access to uploaded files?

Hi @troyeccles,

That’s not impossible, but that’s quite difficult. I don’t remember if you did or if we talked about it, but you should absolutely read my tutorial about training an AI, it’s here. I think you’ll understand better how AI works, and what it really means to feed an AI with data. You would need to constantly train and re-train an AI, or find another way to train an AI once that will be able to extract information in the way you need it from new “content”. Sorry, my explanation here is a bit meta, but the reality is quite complex (a team would need to work on this, basically).

I know many people still think that we just can upload documents to an AI and that it will swallow it happily ?? But the reality is that we still need to train the AI models like children, using huge amount of examples with variations and repetitions.

Hello, to complete the question, is it possible that the AI Engine read attached files? Ex: a word document uploaded (as *.docx file) on the website? Seems complicated but who knows? (you maybe!) ??

Hi @highdem! It’s complicated ?? In fact, I am getting a lot of answers around the idea of feeding an AI model with data, and I am trying to find a way to reply to this and explain it well, but it’s not easy.

Yesterday, I wrote this: https://meowapps.com/ai-engine/faq/#about-ai. It’s not about reading files, but it’s similar (it’s about reading what’s on your website, for instance, knowing your products, etc). Does it make sense when you read this article?

Thanks for the link! Yes it makes sense, but one thing is feeding the model, another is user conversation with the chatbot about the content of a given document. In my view, modeling is about generate answers for every users, while my point about “reading files” is more about having a user discuss with the chat about a given file (for example that he just submitted, or that he reads).

I’m brainstorming and I think one workaround (i’m not asking for that feature no worries!) may be to automatically copy the text of an uploaded document into the body of the {content} but with something like visibility=false so the user will not see the content but the bot will…

Indeed another workaround could be to automatically inject each uploaded document content into the model using a script… If you know a code that does that let us know!

Thank you again for your comments!

I’m brainstorming and I think one workaround (i’m not asking for that feature no worries!) may be to automatically copy the text of an uploaded document into the body of the {content} but with something like visibility=false so the user will not see the content but the bot will…

The problem with that is that you file would have to be smaller than a certain size too, in term of tokens. It’s the problem of being content-aware, the maximum tokens is currently 4000 with davinci :/