Claude for Open Source

[netamismb]

 

[the.rampage.rado]

NOT A MAINTAINER, JUST SPAM MODERATOR HERE.

Yesterday I started testing it. After two PRs in code I have used 50% of my session quota. No matter how better is than GPT if I can use it only when they want I will not use it.

(I'm on the first paid plan, that have 'normal' usage quotas)

[DRMasterChief]

All the free AI offerings have a daily usage limit (or they even fake model versions.). Nothing is truly free if you're looking at cutting-edge models.

However, http://poe.com/ offers some pretty good models for free.

[wakabayashi]

I never worked with claude yet. It seems to be the best model. But right now I can use Codex 5.3 with no limits in phpstorm. I only pay the 20$ plan. Codex 5.3 is very strong as well. Maybe there are "political" reason to leave OpenAI but pricing and quality aren't an issue for me right now 😅

[Rhapsody]

I have been using the free versions of Google Gemini and Microsoft Co-Pilot and am somewhat impressed.  Since I'm retired and just do some stuff as a hobby I don't do enough development work to justify paying for a plan.  I prefer Gemini since it can "see" the tabs I'm working in.  Sometimes it gets stuck in a circular loop that makes the same mistakes when I am trying to debug an issue or it breaks something the was working when it makes a change, and will make the same mistake more than once. In general it is much more efficient than I am.   I used Co-pilot to help Gemini solve one of my circular loop debug efforts and it corrected Gemini on the first try.  I feed that output back to Gemini and moved on further in my regression testing. I copy and paste between the AI and my source as it has no direct connection to the source.

I gave Gemini a set of ground rules and frequently tell it to go back and redo complying with the rules, otherwise it starts changing stuff I did not want touched. Here are some rules I gave it to adhere to when generating or updating my code:
 

The mandatory ground rules found in the testing document are as follows:

• 1. Preservation of Comments: Preserve any comments or annotations in the existing code.

• 2. No Unauthorized Optimization: Do not optimize or change any existing code. Suggestions can be made for acceptance or rejection.

• 3. Patch Formatting: When providing a patch, specify the line before and after the patch for easy location. Use /** START xyz Change **/ as the first line and /** END xyz Change **/ as the last line for inserted code. These may be removed during final cleanup.

• 4. Incremental Updates: Perform updates one function at a time and ask for permission to proceed to the next item to allow for testing.

• 5. Source Code Requests: Request the source for whatever is being updated if it was not already pasted. Use that source for the update unless instructed otherwise.

• 6. Debug Output Encapsulation: All debug output support, which may be done in a production environment, must be encapsulated using the cm_is_super_admin() function to isolate the display. An example provided is:

PHP

/** START DEBUG ENCAPSULATION **/
if (cm_is_super_admin()) {
    /*** DEBUG PLACEHOLDER BELOW ***/
    echo "<br>". __LINE__ . " data:<pre>" . print_r(array_keys($data), true) . "</pre>"; //DEBUG
    /*** DEBUG PLACEHOLDER ABOVE ***/
}
/** END DEBUG ENCAPSULATION **/