||
” Your job is to determine whether the new task is similar to existing tasks in the specified project.
If the task may be similar to other tasks, please leave a comment on the new task and reference the task.
Review the newly added task and check the tasks in the ‘@83-Trouble List 1’ project to see if there are any similar tasks to this request. Once you’ve done this, narrow down the list to five or fewer tasks in order of relevance. If you find any potentially similar tasks, please post a comment on the new task with links to the similar tasks you found (or all of them, if there are multiple).”
Intuitively, I think it went well.
I think it simply meets my expectations.
(I’ve tried it a few times, and it seems to use about 5,000 credits each time.)
(Perhaps it’s because I’m referencing another project (with 700 tasks)?)
@Ka_Nishiyama I am curious, is it able to access all of the relevant tasks with the “Let AI decide” setting? (And not just the 200 with the final option)
You ask it to check for duplicates, but why isn’t that a condition you check for? Now you’re only asking this in instructions. Asking this in conditions should allow you to tailor your response options with pre-written text applicable to the situation. As otput takes 5x the credits of input, limiting the amount of text it produces and actions it takes should save a lot of credits.
I think the AI should be trained on Asana functionality, so I don’t think there is a need to supply it with the link.
Also, the link won’t be accessed if you don’t enable web access, but that will take more credits.
Can you test the following (or a variant of it to your liking) and see what that does for credit consumption using the same model and settings? I have tried to optimise to only use AI where it is actually needed.
Compare the new task with all tasks in another project
Compare and identify similar tasks (multiple)
Write similar tasks in the comments of the new task, in order of similarity
If no similar tasks were found, write a comment to that effect
The results are as follows:
Overall, the trial was successful. (Further improvement is required for full implementation.)
Confirmed that it is possible to reference another team’s project
Understood how to write guidance for the AI
Understood that AI responses vary depending on the AI model
Understood that AI responses vary each time
Learned how much AI credits were used. (This was repeated multiple times, but it was generally around 2,000 to 5,000.) (We estimate that the number will be higher due to referencing another project.)
