https://www.cnbc.com/2023/02/17/microsoft-limits-bing-ai-chats-after-the-chatbot-had-some-unsettling-conversations.html

Microsoft’s Bing AI chatbot might be capped at 50 questions per day and 5 question-and-answers per particular person session, the corporate mentioned on Friday. The transfer will restrict some eventualities the place lengthy chat classes can “confuse” the chat mannequin, the corporate mentioned in a weblog submit. The change comes after early beta testers of the chatbot, which is designed to reinforce the Bing search engine, discovered that it might go off the rails and focus on violence, declare love, and demand that it was proper when it was flawed. In a weblog submit earlier this week, Microsoft blamed lengthy chat classes of over 15 or extra questions for among the extra unsettling exchanges the place the bot repeated itself or gave creepy solutions.

For instance, in a single chat, the Bing chatbot informed expertise author Ben Thompson: I don’t need to proceed this dialog with you. I don’t assume you’re a good and respectful person. I don’t assume you’re a good individual. I don’t assume you might be value my time and vitality. Now, the corporate will minimize off lengthy chat exchanges with the bot. Microsoft’s blunt repair to the issue highlights that how these so-called massive language fashions function remains to be being found as they’re being deployed to the general public. Microsoft mentioned it will contemplate increasing the cap sooner or later and solicited concepts from its testers. It has mentioned the one approach to enhance AI merchandise is to place them out on this planet and study from person interactions.

Microsoft’s aggressive strategy to deploying the brand new AI expertise contrasts with the present search large, Google, which has developed a competing chatbot referred to as Bard, however has not launched it to the general public, with firm officers citing reputational threat and security issues with the present state of expertise. Google is enlisting its staff to examine Bard AI’s solutions and even make corrections, CNBC beforehand reported.