Being generous on ChatGPT allows you to get more detailed answers


A blogger, Thebes, had a pretty interesting idea: see if promising a tip to ChatGPT gets more detailed answers. Result ? That seems to play a role.

Source: Unsplash

Blogger Thebes conducted an interesting experiment to understand the influence of tipping promises on ChatGPT responses.

When informing ChatGPT of a possible tip, its study revealed that the responses obtained were more detailed. The experimental results, published on the X platform (Twitter), show significant variations in the length of the responses provided by ChatGPT.

Source: Thebes on X / Twitter

When the chatbot was informed that no tips would be given, the response was slightly shorter than normal, with a 2% reduction in character count. Conversely, the promise of a tip of $20 (approximately €17) increased the length of the response by 6%, while a promise of $200 (approximately €170) increased it by 11%. To obtain statistically reliable results, the question was asked several times, in order to constitute a sufficiently large sample.

For this experiment, the blogger used the version GPT-4-1106-Preview by ChatGPT. The specific request was to show simple ConvNet code using PyTorch. A simple ConvNet code with PyTorch is a program in Python that creates a neural network to process images. It uses the PyTorch library to define and train this network efficiently.

Longer responses elicited with the tip promise seemed to provide more detailed explanations or add additional relevant content, rather than being limited to a direct response.

Why does this work?

If we query OpenAI, it is explained that ChatGPT is not programmed to react to tips. He responds based on the request, not the money promised. The observed differences could just be due to chance or the way the chatbot interprets the questions. The observed variation in response length could be explained by random variables in the algorithm or coincidence.

But, according to this study, politeness and generosity seem to be essential on ChatGPT. In any case, this observation is particularly captivating when it comes to the training data and its weighting, suggesting deeper implications than might initially appear.


Want to join a community of enthusiasts? Our Discord welcomes you, it is a place of mutual help and passion around tech.





Source link -102