OpenAI has announced that ChatGPT will now remind users to take breaks if they are talking to an AI for too long. The new feature is part of OpenAI’s ongoing efforts to get users to develop a healthier relationship with the often compliant and overly encouraging AI assistant.
The company said in a statement that “gentle reminders” will appear as pop-ups in chats that users will have to click or tap to continue using ChatGPT. “Just checking to see if you’re logged in,” reads an example of a pop-up from OpenAI. “You’ve been chatting for quite some time now – is it time to take a break?” The system is reminiscent of the reminders that some Nintendo Wii and Switch games show if you’ve been playing for a long period of time, though unfortunately there is a darker context to ChatGPT’s feature.
As reported by The New York Times in June, the “yes, and” quality of OpenAI’s AI and its ability to hallucinate factually incorrect or dangerous answers has led users to take darker paths, including suicidal thoughts. Some of the users whose delusions ChatGPT indulged in already had mental illnesses, but the chatbot still did a poor job of consistently stopping unhealthy conversations. In a blog post, OpenAI acknowledges some of these shortcomings and states that ChatGPT will be updated in the future to be more responsive to “high-stakes personal decisions.” Instead of providing a straightforward answer, the company says the chatbot will help users think through issues, suggest questions, and list the pros and cons.
OpenAI obviously wants ChatGPT to be helpful, encouraging, and enjoyable to use, but it’s easy to put those qualities into a sycophantic artificial intelligence. In April, the company was forced to roll back an update to ChatGPT, which resulted in the chatbot’s irritating and overly nice responses. Taking breaks from ChatGPT – and letting the artificial intelligence do things without your active involvement – will make such problems less noticeable. Or, at the very least, it will give users time to check whether the answers provided by ChatGPT are correct.
