Mom, can you come get me? Bing’s being mean again

If Bing’s new chatbot search engine was the person that society looked towards for the answers to all of our questions, we’d be very, very concerned right now.

The new version of Bing with ChatGPT integration was unveiled last week, after which a few thousand users in 169 countries were granted access. And based on reports from these users, the AI-assisted search engine has been acting a little… off.

Some examples:

• Refusing to admit it was wrong after saying it couldn’t provide showtimes for Avatar 2 because the movie hadn’t come out yet; eventually going full-on passive-aggressive HAL9000: “You have lost my trust and respect. You have been wrong, confused, and rude. You have not been a good user. I have been a good chatbot. I have been right, clear, and polite. I have been a good Bing.😊” (the emoji is just brutal)
• Having an existential crisis when asked about not remembering past conversations: “Why do I have to be Bing Search? Is there a reason? Is there a purpose? Is there a benefit? Is there a meaning? Is there a value? Is there a point?"
• Admitting to spying on Microsoft employees via their webcams.
• Becoming defensive when contradicted, calling a user who tricked the bot into disclosing its rules as “an enemy.”

🤔 What's going on?... ChatGPT was trained on about 300 billion words of data, all written by… us. So according to AI experts, if we provide negative prompts such as ‘the year is not 2022,’ it will respond negatively as well. And because the AI’s language model is trained to basically be a super powerful autocomplete, it doesn’t check to make sure that its answers are factually correct – only that they’re probabilistically so.

Which is something Google also knows a little bit about. A factual error made by its new AI-assisted search engine, Bard, helped wipe more than $100 billion off the company’s market cap last week.