Sam Altman says developing ChatGPT is like researching an atomic bomb

Sam Altman expressed concern, comparing the next version of ChatGPT to the Manhattan Project, the nuclear weapons development program that created the world's first atomic bomb.

Sam Altman describes GPT-5 in the style of a horror movie. Photo: Shutterstock .

OpenAI director Sam Altman paints a portrait of GPT-5 that resembles a horror movie more than a product launch, warning that AI is advancing too quickly without adequate oversight.

In a recent episode of the This Past Weekend with Theo Von podcast, Sam Altman describes the AI modeling experience in a tone of suspense that evokes more skepticism than alarm.

He said that GPT-5 “felt very fast,” and recounted moments of feeling very nervous. Despite being the driving force behind GPT-5’s development, Altman asserted that during some discussions, he considered GPT-5 and compared it to the Manhattan Project.

Altman also offered a scathing assessment of current AI governance, with the OpenAI CEO saying that “there are no adults in the room” and that oversight structures are lagging behind the development of artificial intelligence.

As TechRadar notes, the move by the people behind ChatGPT is odd. When selling a product that promises breakthroughs, it's one thing to point out potential risks, but portraying OpenAI as having no control over GPT-5's capabilities is a bit disingenuous.

During the conversation, Sam Altman was vaguely concerned about GPT-5.

What frightened Altman is not entirely clear. He does not go into technical details. His reference to the Manhattan Project is interpreted as an overstatement.

This isn’t the first time Altman has publicly acknowledged his concerns about the AI arms race. He has said that AI can be “completely wrong” and that OpenAI must act responsibly while still delivering useful products.

While GPT-5 will almost certainly come with better tools, a friendlier interface, and a flashier logo, the core issue surrounding this model remains power.

According to Bleepingcomputer 's own sources, GPT-5 could be released in the first week of August, which is similar to previous information from The Verge . GPT-5 is said to be a "unified" model, combining breakthroughs from the theoretical model and multi-modal models, such as o3 and 4o.

Source: https://znews.vn/sam-altman-noi-phat-trien-chatgpt-nhu-nghien-cuu-bom-nguyen-tu-post1572536.html