OpenAI: We're working to reduce ChatGPT bias and bad behavior

February 17 news, OpenAI, the AI ​​research company behind the popular chatbot ChatGPT, said it is working to reduce bias and bad behavior in the system and Users will be allowed to customize some of the chatbot's behavior. Previously, there were numerous reports of ChatGPT’s inappropriate interactions with users and erroneous results.

OpenAI acknowledged in a blog post posted on its company website: "Since the launch of ChatGPT, users have shared a number of outputs that they consider biased, offensive, or otherwise objectionable. In many cases , we acknowledge these concerns, they reveal real limitations of our system, and we want to address them."

The company continued: "We are investing in research and engineering to reduce ChatGPT's There are overt and subtle biases in responding to different inputs. In some cases, ChatGPT currently rejects output that it shouldn't. In other cases, it doesn't reject when it should."

## The #OpenAI move comes in response to reports of bias, inaccurate results and inappropriate behavior by ChatGPT itself, as well as broader criticism from Microsoft and Google of new AI search products currently being tested. In a blog post published on Wednesday, Microsoft reportedly detailed the limitations of its new Bing chat tool based on OpenAI technology, while Google has asked employees to devote more time to manually improving the answers to its Bard system.

OpenAI also said that the company is increasing the development of ChatGPT, including allowing each user limited customization to suit their taste, style and view.

OpenAI wrote in a blog post: "We believe that AI should be a tool that is useful to everyone, so everyone should be able to customize it within the limits dictated by society. This will mean allowing others to System outputs that (including ourselves) may strongly object to. Striking the right balance in taking customization to the extreme may lead to malicious uses of our technology, or the creation of 'sycophant AI' that blindly amplifies people's existing beliefs. is a big challenge. Therefore, system behavior always needs to be limited."

The above is the detailed content of OpenAI: We're working to reduce ChatGPT bias and bad behavior. For more information, please follow other related articles on the PHP Chinese website!