Altman Forecasts New Breakthroughs with GPT-5

If you're tired of the hype around generative AI, you're not alone. OpenAI CEO Sam Altman said in an interview with Lex Fridman on Monday that their latest and most powerful AI model, GPT-4, is actually "kind of bad." In fact, Altman hopes that GPT-5 will live up to expectations.

When asked about GPT-4 and its most impressive capabilities, Altman said, "I think it's kind of bad. I think our job is to look a few years into the future and remember that the tools we have now will look kind of bad in retrospect, and that's how we make sure the future gets better."

Fridman, who had just praised GPT-4, was taken aback by Altman's comments. Altman used this question to discuss his expectations for GPT-5 or the next model, whatever it may be called. The CEO of OpenAI pointed out that his company will release a new version of ChatGPT next year, but did not specify what it will be called. Altman stated that this new model will make significant progress, just like GPT-4.

"I don't want to underestimate the achievements of GPT-4, but I also don't want to overstate them," Altman said. "I think we're on an exponential growth curve right now, and we'll soon look back at GPT-4 the way we look back at GPT-3."

Altman voiced the fatigue many of us feel with AI. We've all heard countless stories about AI taking our jobs, changing the world, or possibly accelerating our impending doom. He described the "amazing flashes" in GPT-4, but pointed out that its best use cases are often as a brainstorming partner. The CEO of OpenAI noted that ChatGPT is rarely useful in handling complex, multi-step problems. When it does work, it's like magic, but those situations are rare.

There has been much speculation about OpenAI's next model. On X, people try to predict the exact release date of GPT-5, GPT 4.5, or GPT-4.5 Turbo every day. Everyone is waiting for that AI model that will truly make our lives better. People were particularly excited when Bing mysteriously cached search results for GPT-4.5 Turbo on the OpenAI website, even though the company stated that they didn't do it and have no plans to release any new products.

In terms of existing models, we have a clearer understanding of Sora. According to an interview with Chief Technology Officer Mira Murati, Sora is OpenAI's text-to-video generator, which may be released in just a few months. This model may include some level of tasteful nudity and has already garnered attention since February with its realistic videos that have shocked the world.