Hi guys,
I am constructing GPT Agents since long. However, today I was really frustrated to see that the agent is not following my instructions Step-by-Step. To check if I was doing something wrong, I copy the prompt of a good agent that I developed before and created a new agent today with the same prompt, but the answers came dramatically different. It seems that the instructions that the chat is receiving from OpenAI before our prompt are significantly affecting the ability of the agent to follow instructions. They might have changed something there.
Can anyone confirm or disprove my theory with your observations?
Looking forward for feedback.
Regards,
VAGR.