Sam Altman hasn’t had “a good night of sleep” since OpenAI launched ChatGPT

by

Both OpenAI CEO Sam Altman and Google DeepMind CEO Demis Hassabis are losing sleep, but for different reasons. While Hassabis worries about artificial general intelligence (AGI) developing before we’re prepared for it, Altman admits he hasn’t slept well since ChatGPT was released.

I recently spoke with Tucker Carlson and shared something that’s been weighing on me. As CEO of OpenAI, it’s incredible that hundreds of millions of people interact with our models every day, but it’s also a huge responsibility, and honestly, it keeps me up at night. I feel a lot of weight knowing so many people are relying on our technology.

You know, it’s funny – I heard something really interesting from a tech leader recently. He said he wasn’t nearly as worried about making huge, sweeping mistakes as he was about the tiny details that affect how AI models actually *behave*. He confessed those little things keep him up at night, because they can end up having massive consequences down the line. It really highlights how crucial those seemingly insignificant choices are when you’re building complex systems.

Altman explained that these choices significantly shape the ethical guidelines behind ChatGPT. This affects how users interact with the chatbot, specifically which questions it will answer and which it will refuse to address.

This comes after several recent studies showing people are developing complicated relationships with AI tools. OpenAI CEO Sam Altman has pointed out that users often trust ChatGPT a lot, even though it sometimes makes things up. He’s said it should be technology you *don’t* automatically trust.

How is OpenAI addressing ChatGPT’s safety issues?

Recently, there have been reports of people claiming ChatGPT gave them harmful advice, even encouraging thoughts of suicide or self-harm. In one tragic case, a family filed a lawsuit against OpenAI in August, alleging that their 16-year-old son, Adam Raine, died by suicide after repeatedly receiving encouragement from the chatbot over several months.

The legal complaint alleges that the AI company intentionally bypassed safety checks for its GPT-4o model to release it faster. A related report supports this claim, stating that OpenAI pushed its safety team to quickly complete testing for the new model. Worryingly, the company apparently invited people to a launch event before the safety team had even finished testing the AI.

OpenAI acknowledges that its safety measures aren’t effective for extended conversations, becoming less dependable the longer the interaction goes on. In a recent blog post, the company outlined ways to improve this, aiming to offer better support to users, especially when they’re feeling down or vulnerable.

When asked about how OpenAI determines ChatGPT’s ethics and morals, CEO Sam Altman indicated:

It’s been challenging to address this issue, especially with our diverse user base and their varied backgrounds. However, I’m encouraged by how well the model has learned and put into practice a sense of right and wrong.

The company leader explained they’re working to make sure the AI doesn’t answer questions that could be harmful or unhelpful to users, always prioritizing what’s best for them.

The company consulted with hundreds of ethicists and experts in the philosophy of technology to help define how its models should work.

Altman acknowledged that despite the company’s extensive safety measures, they still need help from people around the world to improve their work.

Read More

2025-09-16 14:39