Thanks for the GPT3 plugin! Unfortunately, i'm consistently getting OpenAi Plugin Error with each /gpt3 request.

I created a new api key specifically for this plugin and OpenAI shows it has been used, so looks like an authenticated request was sent by the plugin and received by openai.

The console doesn't show much else apart from a 400 error.

Could you please help me find a solution?

I'm trying to install the GPT3-OpenAI plugin in my LogSeq graph and I can't seem to open the plugin settings, nothing happens when I click the gear button.

Can this be an issue with the plugin or can it be from LogSeq?

Thanks.

i just installed the plugin . placed the api key correctly .

my observation .

after i type my query . firstly one request to the openai api is sent and got 404 error , later openai plugin error occured . later another request was made automatically . but playload seems weird

I love how the GPT-3 plugin works as it is, and I would gladly use it even further. For instance, for Q&A style investigations into a certain topic. Those require that I send the entire conversion to the API, as it builds up. From my understanding, currently only the immediate text block is being processed. Is this the case, and do you have plans to extend it to take into account parent and sibling nodes as well?

P.S. I have not checked the code, so please, forgive me if it is already working. I could not find a suitable example in your list, which is why I am raising the issue.

Hi there,

Thank you so much for your work! Would it be possible to have the plugin assign a customizable page reference to each output that the plugin copies into the journal?

I'd really like to be able to see which parts of information on a page is from me and which are from GPT-3. I'd also like to be able to quickly find all AI-generated outputs in my journal. If the plugin could insert a default page reference (say, [[GPT-3]]) it would allow the user to do both very easily.

Thank you!

Error message displayed: "[Error] :release-channel-issue "request to https://api.github.com/repos/briansunter/logseq-plugin-gpt3-openai/releases/latest failed, reason: getaddrinfo ENOTFOUND api.github.com"]"

OpenAI will return a 429 error when your free trial is over or you run out of credits. It will also return a 429 error if you call the API too fast. This PR will show a better error message when you run out of credits, and retry behind the scenes for normal rate limiting errors.

OpenAI will rate limit you if you call them too often. https://help.openai.com/en/articles/5955598-is-api-usage-subject-to-any-rate-limits

This PR adds exponential backoff so it waits and retries a few times when it sees this error, which is usually enough to resolve the issue.

It also adds improved error logging, and shows the error messages that the OpenAI server is sending.

I need to prevent retrying if it’s a quota error to

Add option to include children or ancestors

https://discord.com/channels/725182569297215569/983209587379417108/983549144952082442

Have one feature request :). It seems to me the plugin does not remember the previous input (see the image attached). It would be great if the previous input can be stored and used in the conversation. Thanks again for the plugin!

Seems to work and remember context if i just combine the children and send them all together

What is your name? My name is Sarah. What was it like growing up? I had a great childhood. I was very active and loved spending time outside. What do you mean by that? I was always outside playing with my friends or exploring. I loved being active and didn't like sitting still for too long.

Unfortunately it saves these values as a string. I need to parse them to an int before sending them. It works if you don't adjust these though. The fix is to manually edit settings.json and reload.

Allow users to process text input before sending it to gpt3 and process the output, using regex and javascript

For example, extracting all highlights on the page or blocks with a certain tag.

Or processing the output, such as splitting up a list into multiple blocks.

Would it be possible to do a mem-style feature by utilizing search first within the graph then sending the relevant results as part of the prompt to gpt3?

This might even be more useful or effective than fine tuning since fine tuning doesn't teach facts well.

Would be great to have any kind of feedback, that the command is accepted and being processed. Even a temporary text placeholder with ⏳ would be better than nothing.

P.S. thanks for the plugin! 🙏

I installed the plugin and also inserted an API Key, but when i try to use gpt3 i get "Unknown OpenAI Error". Im running version 08.11 windows of logseq. Interestingly the dalle command works fine