new extension: power chatgpt, bring chatgpt into the roam

[dragonforce2010]

…l to make our work more efficient!

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt d53299d

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt 9eaa759

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt 2383a38

[panterarocks49]

What is chat-pro-screen and why does it look like compiled code?

[dragonforce2010]

What is chat-pro-screen and why does it look like compiled code?

yes, it's compiled code. Let me optimize this pr.

[baibhavbista]

Hey @dragonforce2010

I saw that this extension sends the GPT requests to https://ec2-54-254-24-184.ap-southeast-1.compute.amazonaws.com/ and after that I assume forwards the request to OpenAI.

I'm afraid we cannot approve this extension while it does requests to an unknown backend. If you change it such that it directly does a fetch on openAI's servers, that would make it okay.

Also, as @panterarocks49 already mentioned before, we cannot approve the extension while it has compiled code like src/chat-pro-screen/sdk.js

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt 4103abd (PR-shorthand: dragonforce2010+roam-power-chatgpt+529)

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt c473b45 (PR-shorthand: dragonforce2010+roam-power-chatgpt+529)

[dragonforce2010]

Hey @dragonforce2010

I saw that this extension sends the GPT requests to https://ec2-54-254-24-184.ap-southeast-1.compute.amazonaws.com/ and after that I assume forwards the request to OpenAI.

I'm afraid we cannot approve this extension while it does requests to an unknown backend. If you change it such that it directly does a fetch on openAI's servers, that would make it okay.

Also, as @panterarocks49 already mentioned before, we cannot approve the extension while it has compiled code like src/chat-pro-screen/sdk.js

@baibhavbista all the comments have been addressed, please review it, thanks

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt 8e53481 (PR-shorthand: dragonforce2010+roam-power-chatgpt+529)

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt e4f79d2 (PR-shorthand: dragonforce2010+roam-power-chatgpt+529)

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt c5a0da0 (PR-shorthand: dragonforce2010+roam-power-chatgpt+529)

[github-actions]

Here’s your link to the diff:

Added dragonforce2010/roam-power-chatgpt 9cb5e2f (PR-shorthand: dragonforce2010+roam-power-chatgpt+529)

[baibhavbista]

sent some comments via slack @dragonforce2010

[baibhavbista]

Adding my notes for the comments I had here too, for future reference:

• something weird happened when I ran ./build.sh locally. It produce 100+ js files in the directory

  • 

• they'd added an API token of their own. Seems to not be working. Since the token is in a public github repo, would not be surprised if a scraping bot found it. Might be a good idea to remove the default API token and force users to use their own key

  • 

• when sending the "messages" in the chat/completions request, wrong messages are being sent (the first 8 are being sent while the last 8 should be)

  • 
    • here what it’s doing is only keeping the first MaxContentMessageSize messages on each request
      while you probably want to be sending the last MaxContentMessageSize messages

• I think max_tokens should be customizable. Turns out it was has been hardcoded in the source code to 200. I had been wondering why my requests were getting cut off in the middle

  • 

• want to use extensionAPI’s addCommand for the "Roam GPT" command instead of the roamAlphaAPI one

  • 

• nitpick: spelling mistake in “try again”. might be problematic since it is a message sent to openai as well and may give wrong answers

  • 

• I’d also expect most users of Roam are english users (+ best experience when interacting with chatGPT is in english I think) so maybe placeholder text should be in english as well?

  • 

• slight nitpick: what about only showing the parts inside triple quotes using codemirror?

  • 

• another codemirror issue: when a line is highlighted, it’s white by default which is a problem given that font is also mostly white. Compare this with the earlier screenshot

  • 

If I have more comments, will be sending them via slack

[baibhavbista]

@dragonforce2010 reminder that this PR is still in requested changes state

[baibhavbista]

Marking this as a "draft PR" in order to not confuse & accidentally release

Please change back to a regular PR after looking at the requested changes @dragonforce2010