ChatGPT is reshaping the way people use AI

Can ChatGPT and other artificial intelligence tools learn from the input they get and the feedback they receive from users?

Yes! In fact, this was a key process in its creation and is still employed for fine-tuning, shares Ryan Mattfeld, Assistant Professor of Computer Science for Elon University News Bureau.

Large volumes of raw data are the foundation of recent machine-learning technologies. This also applies to ChatGPT, whose initial training was based on substantial quantities of data gathered online. Yet as ChatGPT's developers, OpenAI, note on their website, humans had a significant role in the program's development. Following the creation of the enormous dataset, human labelers adjusted ChatGPT's replies. In the beginning, a person asked a question, and another human gave what they thought would be a suitable response. Another round of fine-tuning started once this information was taken into account. Many distinct answers to the same question were created by the AI in this stage, and human labelers then graded the AI solutions from best to worst. This criticism was also implemented into the network.

You may now choose whether to give the replies that ChatGPT generates a "thumbs up" or a "thumbs down" while using it yourself. If you select "thumbs down," you may explain why you thought the response was unsatisfactory. This is applied to improve ChatGPT even more.