The discussion revolves around debugging AI personas in Discourse, particularly when they fail to respond consistently despite being properly configured. Overgrow reports that their AI bots, created using the new AI Persona Editor and powered by OpenAI’s 4o-mini model, sometimes fail to reply when tagged, with no errors visible in logs. They express concern over the unreliability, which is affecting user experience. Falco asks about the LLM used, and Overgrow confirms it’s OpenAI 4o-mini. Saif suggests checking the persona settings like Allow topic mentions and Allow chat channel mentions, and speculates that the issue might be related to the prompt design, highlighting the need for a more structured debugging process. The thread seeks guidance on identifying the root cause of inconsistent behavior and improving reliability.
I’ve created a custom persona/bot using the howto. After some initial success, it seems to work sporadically. I have yet to identify the root cause. The problem is that no reply is created by the bot, and there is no sign of any error in /logs.
What is the next step to try or look for?
How do you generally debug AI personas?
I’ve also noticed that while the information icon (for debug purposes) is properly shown under each bot’s post when interacting in a PM with the AI, there is no such icon in their public posts. Is that correct?
Sorry to bug you again. I don’t mean to add more work for you, but I’m wondering about the proper method for debugging AI personas.
I’ve followed the guide and created two bots. The issue is that they sometimes react when tagged, and sometimes they don’t. Even with the same user and privileges. It seems to happen randomly, and no errors appear in the log. Is there any debugging process available in the current version that I could use?
It’s getting to the point where I’ll probably have to disable them because users keep making fun of the bot’s unreliability.