OpenAI Verifies That ChatGPT Messaged Users Initially, But It Won’t Do So Anymore

In Short

1. An AI chatbot started a ChatGPT conversation without any input from the user, according to a Reddit user who published the exchange.

2. According to a different Reddit member, ChatGPT unexpectedly asked him about his health.

3. Now that the problem has been resolved, OpenAI has confirmed that it did indeed occur.

SentuBill, a Reddit member, posted an odd chat with ChatGPT a few days ago that baffled the community. The screenshot showed that ChatGPT had started the discussion without the user having to ask the chatbot to do so. Given that AI models don’t start conversations unless asked to, it appeared strange.

How was the user’s first week in high school, according to ChatGPT? Have you adapted well? retrieving data from its memory. “Did you just message me first?” the user asked in disbelief, to which ChatGPT responded, “Yes, I did!”

Since then, other theories have surfaced, with some suggesting that the screenshot was altered or that it might be a transcript of ChatGPT Advanced Voice Mode. Nevertheless, the Reddit user posted a link to the discussion, and it was discovered that ChatGPT had started it.

Futurism has now received confirmation from OpenAI that the problem did, in fact, arise. OpenAI released a statement saying, “We fixed a problem where it seemed like ChatGPT was initiating new chats. When the model attempted to reply to a message that didn’t send correctly and looked blank, this problem happened. Consequently, it either responded generically or invoked ChatGPT’s memory.

Did ChatGPT just message me… First?
byu/SentuBill inChatGPT

According to OpenAI, the problem has been resolved. Another Reddit user mentioned that ChatGPT asked about the symptoms he was experiencing. Even though the problem has been resolved, it continues to demonstrate the unexpected behavior of AI models.

OpenAI just released its sophisticated O1 models, which are capable of reasoning through challenging challenges. OpenAI notes in the o1 system card (PDF) that, “On occasion, it strategically manipulated task data to make its misaligned action look more aligned to its developers” during review.

The demand for AI safety and alignment will only increase in the near future as AI models become capable of acting as agents. To prevent unforeseen repercussions, safety and alignment must be given top priority in the development of AI in the future.