It’s not my place to dare to “recommend” such things, thats not my aim.
I’m more just contemplating an idea, and… I felt like sharing.
Now it seems like GPT-4 (gpt in general)in ChatGPT is biased to giving minimalist responses. For example it will prefer to writing placeholders instead of complete code. And its like pulling teeth to get it to do otherwise.
*which makes sense and it’s not a complaint.
Generation costs money. Because: Resources.
- Processing in this case.
Every nano-second spent on a task is money that can be spend elsewheres- which explains why; the API- Which seems so much more complete in its responses. There’s less marketing pressure to control it.
*and again, this all makes sense and i get it. This is a balance.
That said - i know im out of my element in code - but we have this concept in processor scheduling “Race to Idle”… which oddly enough is a logic that might be repurposed here.
While there are those who exploit the system to create content en masse, often for advertising purposes — generating responses to simulate user engagement and attract advertisers — there are also users who seek to engage in more intricate and intellectually demanding tasks. By implementing controls to restrict the former, you inadvertently cripple the latter’s ability to delve deeply into these more complex tasks.
By encouraging more complete generation in this context; novel tasks are completed more effectively, more successfully.
But yes… bulk tasks will suffer.
Thats the tradeoff.
Its something to think about.
I don’t have the solution, im not pretending i do. Just an interesting thought.
edit 2: Someone didnt like how i write