User talk:Phlsph7/WikiChatbot

Error when using

I recently tried using this tool, and I'm getting error Error: The error code is 429. This indicates that you have sent requests too quickly or that you have reached your monthly limit.

I made a new API key for this, and this is an API I have never used, so I don't really understand how I reached my monthly limit if I've never used it prior? Like I've used ChatGPT, but that surely doesn't have anything to do with it.

Cheers, --QuickQuokka ^{[⁠talk • contribs]} 19:14, 7 September 2023 (UTC)[reply]

Oh I may be stupid. The API is paid.

Respectfully, your tool seems very cool and I wish I could use it, but I don't really want to spend money on something that is not necessary to me. QuickQuokka ^{[⁠talk • contribs]} 19:22, 7 September 2023 (UTC)[reply]

Hello QuickQuokka and thanks for letting me know. As I understand it, the API is free to use for new accounts: the free period starts when the OpenAI account is created. So if you created your openAI account more than 3 months ago then the free period would be over by now. This was the case for me: when I started experimenting with API keys before I wrote this script, my account was already older than 3 months. I was unable to use them for free and I had to add a payment method. For this reason, I could not test for myself how the free period for the API works. In case you want to give it a try with the paid version, you probably need a credit card. The costs are relatively low, usually a fraction of a cent per query. Phlsph7 (talk) 19:40, 7 September 2023 (UTC)[reply]

Using GPT4-32k to draft full articles from at least 5 RS

Many thanks for putting this together. Following up to my comment on Wikipedia_talk:Large_language_models, I've recently had success drafting full wikipedia articles by feeding the text of up to 5 RS (so far) into GT4-32k through openrouter.com/playground and then giving it the right prompt. I first use a text program to assemble the contents of the RS in the following format:

Reference 1:

(put url of reference in here)

(cut and paste web page as text here)

Reference 2:

etc.

The assembled text then gets cut and pasted into the system prompt of GPT4. A prompt to draft the article is then pasted into the chat and GPT4 drafts the article. Section headings and references are properly formatted. Here's an example of a chat prompt:

Write a Wikipedia biography for Harrison Floyd based on Reference 1, Reference 2, Reference 3, Reference 4, and Reference 5. Use all of the references. Use only information found in the references. Use ref name format for the inline citations to the references. Include the urls and details of the references in the citations.

The output can be found here: [1]

Would be happy to help work on a GUI where I could just drop in urls instead of cutting and pasting. Nowa (talk) 00:19, 20 September 2023 (UTC)[reply]

The idea is interesting and implementing this into a script wouldn't be too difficult. However, there are major obstacles to ensure that the result complies with Wikipedia policies. See the discussion at Talk:Harrison_Floyd#Initial_content_summarized_from_references_using_GPT4. Phlsph7 (talk) 07:47, 20 September 2023 (UTC)[reply]

To be more precise, there would be a problem of automatically getting the text from the article urls: CORS prohibits client-side scripts from accessing other websites. I don't think it wouldn't be possible to implement this feature with a regular user script and people would still have to copy-paste the texts of the sources. Phlsph7 (talk) 08:07, 20 September 2023 (UTC)[reply]