OpenAI, the startup behind the viral chatbot ChatGPT, is working on an upgrade that will allow users to customize the technology. The goal of the upgrade is to address concerns about bias in AI and to accommodate more diverse views.
The San Francisco-based startup, which is funded by Microsoft and powers its latest technology, is working to mitigate political and other biases while still allowing for some disagreement in system outputs. The upgrade will offer customization as a way forward while still setting some bounds on system behavior.
ChatGPT, which uses generative AI to produce answers mimicking human speech, has sparked frenzied interest since its release in November 2022. However, some media outlets have raised concerns about the technology, such as potentially dangerous answers from Microsoft’s new Bing search engine, which is powered by OpenAI.
Companies in the generative AI space are focusing on setting guardrails for this nascent technology, and Microsoft has said that user feedback is helping it improve Bing before a wider rollout. OpenAI’s guidelines for ChatGPT aim to ensure that the technology remains responsible and accountable.
The answers generated by ChatGPT are first trained on large text datasets available on the Internet. As a second step, human reviewers review a smaller dataset and are given guidelines for different situations. In cases where the user requests adult, violent, or hate speech content, ChatGPT will respond with “I can’t answer that.” For controversial topics, ChatGPT will offer to describe viewpoints of people and movements, rather than trying to take a correct viewpoint.
The upgrade to ChatGPT is part of OpenAI’s ongoing efforts to improve AI and ensure that it benefits everyone. The startup is working to develop technology that is unbiased, transparent, and accountable, and that can be used to address some of the world’s most pressing challenges.
For the latest tech news, follow Techbar Nepal on Facebook, and Instagram.
For Digital Products and Services: Maurisys Software.
