LLM Punisher: Refund Me When AI Messes Up!

Discussion
1

Pain turned into a joke or an idea: create an “LLM Punisher”—a tool that automatically refunds the API costs whenever an LLM fails to deliver the promised output.

Here’s the basic idea:

  • You prompt the LLM.
  • It provides an answer (and you pay for it).
  • If the answer is incorrect, you send it back for re-check.
  • When the LLM admits its mistake, you’re refunded the cost of that prompt.

In short: Pay for accuracy, get your money back for hallucinations.

Thoughts?

  • No way
    or
  • YES PLEASE! :robot:
10 Likes
2

you are right .

1 Like
4

Thought is valid but need to rationalize it. I feel, this is not that simple.

5

Easy refund that would model dumb and always responded wrong done = endless requests = loss of funds cursor

6

In a perfect world, it would be great. Though I doubt it will be introduced, unless Cursor wants to be nice, since it costs Cursor the same whether the code output is correct or incorrect.

One thing I noticed that should be rectified (haven’t checked recently if it has) – you get charged for the message as soon as AI starts responding, but if an error occurs midway that stops the AI halfway, you still pay for the full message, and then another to finish the halfway halted message.

3 Likes
7

It’s not very realistic, whether successful or not Cursor will have to bear the cost of accessing LLM unless the pricing of the correct answer includes those costs, and even then it’s unlikely because there will always be people abusing to ask impractical questions

8

That could be a cool idea for a time-limited promotion tho.

  1. turn it on
  2. you gather feedback on what to improve
  3. turn off
  4. improve software

Ofc there need to be limits for bad actors.

1 Like