Problem with Content Aware

Hi @mrgoldberg,

The problem is that often, the content of the page is too big; if I don’t resize it, it will make the conversation very limited (or it will not work).

Currently, I am keeping the 1024 first characters of the content (while avoiding to cut a sentence right in the middle). I am going to update this to 1024 tokens soon, as it makes a bit more sense.

Isn’t this what you are getting?

There is really a lot of content on the page, as there are reviews of 10 product models. The task that I wanted to solve was for the chatbot to be able to recommend models from this page, and not the models that it has in the database.

Yeah, the embeddings are not always based on your post/pages. They are just pieces of content.

So you could create an embedding per product for example. A few more embeddings that compare a few products together, or about a technical aspect at our product share.

For example, if you are Apple, I would create an embedding for each iPhone version, with the details about each iPhone, then an embedding about the differences between iPhone and Samsung and others, an embedding about why iOS is great and what are its features, etc, etc.

There is a browser plugin like https://github.com/qunash/chatgpt-advanced. It allows you to collect content from the page completely, or even from several. Maybe you can implement something similar but only in your plugin? Ideally, the bot should only communicate using the content of the current page.