which are controlled by other chatbots

Study suggests AI chatbots can rate each other

Unreliable responses are one of the big problems with AI chatbots. These tools are developing by leaps and bounds, but they are still they reproduce prejudices and give rise to hallucinations. Researchers at the University of Oxford came up with a special idea to overcome these limitations: chatbots are controlled by other chatbots.

The group, consisting of Sebastian Farquhar, Yannick Caussin, Lorenz Kuhn and Yarin Gal, notes that false results have hampered the adoption of chatbots in various fields. Now, he says, the method they’ve developed takes into account the fact that an idea can be expressed in a variety of ways, and allows users to identify when in a conversation they need to be especially careful.

Using chatbots to monitor other chatbots

The researchers asked the chatbot a series of simple questions and math problems. Then they asked a group of people and another chatbot I’ll consider the answers. After purchasing ratings, they found that the chatbot had 93% agreement with human raters. Human evaluators, for their part, agreed with the chatbot by 92%.

These results are part of a study published in the journal Nature entitled “Detecting Hallucinations in Large Language Models Using Semantic Entropy.” As we can see, a fairly manual methodology was used, which can be used as a model to inspire possible automated solutions to address the unreliability of the AI ​​chatbots we use every day.

The tools themselves often include warning messages about the accuracy of responses at the bottom of the chat window. “ChatGPT I can make mistakes. Consider checking important information,” says the OpenAI chatbot. “Gemini may display inaccurate information even about people, so double-check their answers,” reads the suggestion created by Google.

Tango OpenAI, like Google and other companies, have said they are working to improve the reliability and security of their artificial intelligence products. However, at the moment the results are far from perfect. In many cases, the answers are usually text that appears very coherent at first glance, but may contain everything from minor inaccuracies to major errors.

Images | Xataka with Bing Image Creator

In Hatak | Artificial intelligence created a film script that specifically talks about creativity in cinema. The theater refuses to release it