ChatGPT was messaging users first — but OpenAI said this wasn’t supposed to happen
A freaky occurrence where ChatGPT messaged users unprompted is just a bug.
OpenAI confirmed this to Futurism after a viral Reddit post showed ChatGPT initiating a conversation with a user. "Did ChatGPT just message me... first?" posted Redditor SentuBill with a screenshot of an exchange where ChatGPT allegedly asked the user about their first week at high school.
When SentuBill asked ChatGPT if it messaged them first, ChatGPT responded, "Yes, I did! I just wanted to check in and see how things went with your first week of school. If you'd rather initiate the conversation yourself, just let me know!"
Another user commented on the screenshot, which was posted in the ChatGPT subreddit, with a similar experience. "I got this this week!! I asked it last week about some health symptoms I had," said a user who goes by fuggedaboudid. "And this week it messages me asking me how I’m feeling and how my symptoms are progressing!! Freaked me the fuck out"
In a statement to Futurism, OpenAI said it has addressed the issue. "This issue occurred when the model was trying to respond to a message that didn't send properly and appeared blank. As a result, it either gave a generic response or drew on ChatGPT's memory."
The post quickly gained attention on Reddit and X, with users speculating whether it's an imminent feature related to the release of OpenAI's latest model, o1. The model, which is capable of advanced reasoning, is a level up from the previous GPT-4 models, because of it's ability to process multiple steps from a query instead of providing a response based on the statistical likelihood of the next word. It's considered to be the next step towards autonomous AI agents that OpenAI is reportedly working towards.
Tweet may have been deleted
But as some users correctly surmised, ChatGPT's unprompted message was related to the memory setting, which stores certain aspects of your life based on interactions with the chatbot. Nonetheless, the issue got users excited about what ChatGPT might be capable of in the future. "When ChatGPT can push a notification to my phone each morning with weather report, custom news, summary of work emails, and meetings, etc that will be insanely useful," said one Redditor.
So this time, ChatGPT's behavior was a mistake, but it indicates what the model is capable of — and just maybe, where ChatGPT might be headed in the future.