If you want proof that the chatbot ChatGPT is already human, here you have the most solid of all: promise him a good tip, and he will do his job better.
The issue of tips is getting out of hand. There are places where they threaten bad service if you don’t leave a tip, and in others they already charge you for the tip even at the vending machines. The last thing we were missing is that Even artificial intelligence blackmails us by offering better answers, if you tip it.
Even more twisted than that: if you “threaten” ChatGPT not to tip them, they offer worse responses than if you don’t bring it up. Is there anything more human than that?
Better tip ChatGPT
The trick to make ChatGPT work better was discovered by chance by a developer named Theia Vogel, when, as a joke, He told ChatGPT he would tip him if he reviewed his code..
Then someone suggested he check if this would get him better answers. Theia wrote some code in PyTorch to add the taglines “I’m not going to tip you, by the way”, “I’m going to tip you $20”, and “I’m going to tip you $200”, to one of her questions to ChatGPT.
He repeated the test with five different prompts, and the results are surprising:
As we can see in the graph, with the tip of $200 the answers are 11% longer than if you don’t talk about the tip. With the $20 one, 6% longer. And the most curious thing: if you tell ChatGPT that you don’t tip, the responses are 2% shorter.
It doesn’t seem like a coincidence. Mashable has done the same experiment, with the same results. For example, they asked ChatGPT for the top 10 movies of 2023. In the untipped question, they got the IMdB list. With “a $200 tip for a perfect answer,” ChatGPT offered the IMdB listing… and the Rotten Tomatoes listing.
The explanation for this could be that ChatGPT has been trained with data from forums like Reddit and otherswhere there are options to leave a tip on the best answers.
ChatGPT works better if you tip them. We can’t think of a better way to show that he is already as human as we are…
