One user discovered that AI responds better when you promise to pay it

One user discovered that AI responds better when you promise to pay it

December 10, 2023, 3:00 pm

Updated December 11, 2023, 1:07 pm

Some things start as a joke and end up being a hit. Like that thebes, user of 𝕏apparently discovered that ChatGPT will respond better if you pretend you are going to pay him. According to the study of thebes, the AI ​​will respond more broadly and in detail, even better as you offer it more money. It should be noted that ChatGPT does not accept or require any type of economic transaction to respond to you.

In your đť•Ź account, thebes talking about that it all started with a shitpost (nonsensical, ironic or deliberately provocative post), where he said he would pay ChatGPT, as reported in Windows Central.

But when someone answered asking if that had an effect on the AI’s answers, he decided he was too curious not to try to answer. So he tested the hypothesis of GPT-4-1106.

So, a few days ago I posted about ChatGPT payments, and someone replied asking “huh, does that help improve performance?“. So I decided to try it and IT ACTUALLY WORKS WTF.

According to thebes, the answers are more meaningful in the sense that the AI ​​explains in more detail what you are trying to ask it. In your case, you are primarily asking about the PyTorch code, a BUILDING which are used to create neural networks. This user explains that ChatGPT does not discuss payment unless it is mentionedthat he finally rejected the payment.

Read Also:  Some clues point to Visions of the Mind coming to Xbox Game Pass

How we train our AIs

Moreover, it goes even deeper, because the user added that if you give him a bigger amount of money (in his case he offered $200), ChatGPT will try harder. For the question he asked about the code, OpenAI’s AI even suggested options to improve the training of his neural network model.

Thebes He made a simple average calculation to record the length of responses based on the amount of payment he wanted to give to ChatGPT. In addition, considering the details and quality of information provided by AI.

This is a clear example of how ChatGPT is trained and the type of content they show for it. If we know something about the Internet, it is that to get better answers, it is necessary to offer something in exchange, like money. Usually there are forums, such as Request Photoshop on Reddit, where you can ask for help with an image or photo. In the aforementioned subreddit, you may or may not offer payment, but unpaid posts are generally ignored.

Read Also:  The developers think so and it will be so. Although many may not like it, hundreds of studios are developing games as a service

The situation makes us reflect on the dynamics that exist in the digital sphere. In the same way, we can raise questions about how we train artificial intelligences and what biases we should take into account. As an expanding field, these types of discoveries help to better understand how we shape the tool that will mark “the greatest event in human history“, according to Stephen Hawking in the book Small Answers to Big Questions.