OpenAI CEO Sam Altman Says Superintelligence Is A Few Thousand Days Away

OpenAI released its new o1 models recently delivering breakthrough performance in complex reasoning and inference scaling. Internally, the ChatGPT maker believes that it has achieved Goal 3 which is to build an AI agent that can perform complex tasks.

Now, in a blog titled ‘The Intelligence Age’, Sam Altman, the OpenAI CEO makes an astounding claim and says superintelligence is a few thousand days away. He writes, “It is possible that we will have superintelligence in a few thousand days (!); it may take longer, but I’m confident we’ll get there.“

Now, a few thousand days might mean five to ten years, but we can’t say for sure. However, in 2023, OpenAI shared its vision on how to govern superintelligent AI systems and said, “it’s conceivable that within the next ten years, AI systems will exceed expert skill level in most domains, and carry out as much productive activity as one of today’s largest corporations.”

Here, OpenAI indicates that by 2033, we may have far more powerful AI agents that can carry out the tasks of an entire corporation. In the blog post, Altman further notes that deep learning works and it will get better with scale and more resources. OpenAI has been bullish on scaling current AI systems and predicts that scaling large AI systems will inevitably lead to intelligence.

Ilya Sutskever, co-founder and former chief scientist of OpenAI, also believes that scaling, both in terms of the size of the neural network and computing resources, may lead to greater intelligence. Sutskever has now left OpenAI to build his own company called Safe Superintelligence Inc. (SSI).

That said, other AI researchers, most notably, François Chollet who works at Google argue that scaling existing technologies like LLMs may not be sufficient to achieve Artificial General Intelligence (AGI). Chollet has even designed an evaluation known as ARC-AGI to assess the generalized intelligence of AI models. By the way, OpenAI o1 models performed poorly on the ARC-AGI benchmark.

Yann LeCun, who is the chief AI scientist at Meta, also argues that LLMs can’t plan and are essentially autocomplete systems rather than world models. Many AI researchers on the other side say that new breakthroughs in AI are needed to unlock superintelligence, and LLMs may not be enough.