OpenAI chief Sam Altman recently addressed an audience at the University of Michigan, sharing his insights on OpenAI’s latest reasoning model and the future of artificial general intelligence (AGI).
“You know we numbered this as one because it is a very early thing,” said Altman, saying how they were very close to naming the model GPT-5 instead of o1.
Further, he said since the model had different purposes and did things that the GPT series didn’t, they took a conscious choice of not naming it GPT-5. “It is good at different things. It takes a long time for hard problems which is annoying, but we will make it better,” he added.
It is interesting to note that in almost every interview, prior to a new launch, Altman downplays the previous model, likely to build hype and anticipation for the next one. In a previous interview, he described the o1 model as in the “GPT-2 phase” of reasoning, noting that while early in development, significant progress is expected soon.
The definitions for AGI are now virtually ‘meaningless’
“I would like to banish the word AGI because it has become so overloaded,” noted Altman, rendering the term meaningless as people have started to associate so many things with it. He also clarified that some even believed GPT-4 would be AGI, but it’s not.
He also noted that in 2019, no one could have imagined a model like o1, and fast forward to five years from now, in 2029, the progress in AI will be just as unbelievable. “We are on this exponential curve, and every year from 2019 to a long time in the future is going to look like breathtaking forward progress,” he said.
In a previous essay, titled the Intelligence Age, Altman mentioned that super intelligence could be a few thousand days away.
GPT-5 Coming Soon?
With the delay, some anticipate that OpenAI may directly skip straight to GPT-5. This could also be possible due to OpenAI focusing on transitioning to the next frontier model infrastructure for its future releases. Their DevDay, earlier this month, saw a series of technical demonstrations and releases, building up to the hype of GPT-5, and hinting at its arrival in the coming months.
Recently, OpenAI launched Swarm, an open source framework for building and deploying multi-agent systems. “Swarm is designed to make agent coordination and execution lightweight, highly controllable, and easy to test,” said the OpenAI solutions team.
Last week, NVIDIA shipped its highly anticipated blackwell AI chips to OpenAI and Microsoft, delivering three times faster training speeds and fifteen times greater inference performance.
All of these developments come in the backdrop of the company raising $6.6 billion in funding, bringing its valuation to $157 billion. With this, the company is looking to accelerate frontier AI research, alongside expanding compute capacity, and continuing to innovate in artificial superintelligence (ASI) development. “We’ll do whatever works to build safe AGI and figure out how to share the benefits,” said Altman.
The post ‘OpenAI’s o1 Model Was Almost Named GPT-5,’ Reveals Sam Altman appeared first on AIM.
