Access external URLs about pokitoki

[HeySlava]

What would be the expected response if a single request to OpenAI exceeds the content limit (a long article)?

[nalgeon]

It would be an error returned by the OpenAI API.

[HeySlava]

I agree. However it's possible to split a text into several requests

[HeySlava]

            I am sorry. I thought about different thing. Something like this: max_tokens=1000

[nalgeon]

The max_tokens parameter in openai.ChatCompletion is the maximum number of tokens to generate (the number of output tokens). It's currently hardcoded to 1000.

The number of input tokens is determined by the user message. After replacing the URLs with their content, it could be significant.

OpenAI has a limit on the total number of tokens (prompt + input + output). It will return an error if there are too many tokens.

All of this is beyond the scope of this issue. Handling large numbers of tokens is important, but that's a question for another time.

[nalgeon]

Here goes: External links

[HeySlava]

    def is_text(self) -> bool:
        """Checks if the content type is plain text."""
        if not self.content_type:
            return False
        if self.content_type.startswith("text/"):
            return True
        if self.content_type in self.allowed_content_types:
            return True
        return False

Isn't it better to raise an exception instead of returning False? My idea is simple: users will understand that their links are unprocessable before request to Openai.

[nalgeon]

I prefer the current implementation. The bot states that it can't process the binary content:

Q: Explain this paper in simple terms https://arxiv.org/pdf/1706.03762.pdf

A: I'm sorry, but I cannot access the contents of the paper you provided as it appears to be unknown binary content. Can you please provide me with the correct link or the title of the paper so I can provide you with a clear explanation?

[HeySlava]

This is my example:

Read the most popular comments: https://www.youtube.com/watch?v=outcGtbnMuQ
I apologize, but the content you provided does not include any comments from viewers. Can you please provide a different link or specify where the comments can be found?

There are not handlers in case when content_type not in self.allowed_content_types but link contains text-ish content

[nalgeon]

This is another case. The content type is OK (text/html), but the actual content is loaded dynamically via JS. I don't think I will support such cases.

[nalgeon]

I'm pretty happy with with the implementation so far.