I’m creating assistants and threads to manage a chat conversation on twitch. This thread takes one of two paths, “respond” or “fact” (share a fact).
I’m seeing two issues:
- Occasionally the random_fact thread will generate a response without receiving any thread instructions
- The GPT outright ignores run instructions
I feel like I may be misunderstanding a core concept about how “Assistant” vs. “[assistant_name]” (generated by my app), “run_instructions” and “user” are working here… and how these interact ordinally.
I’ve attached a screenshot, you can see cases where both ‘random_fact’ responds twice in arow without receivin any run instructions as well as a case where the GPT ignores instructions completely.
Sidequestion: Is it possible to generate a second response from my assistant (random_fact) without providing run instructions? I can’t think of why two would be generated…
Note: I’ve found that including a generic “assistant” message like “(No conversation happenign here…” helps to direct the bot a little, but seems like overkill and still doesn’t prevent the issue from happening.
Feel free to ask any questions, I may not have explained this clearly or could be misunderstanding a core concept of how these work.
