OpenAI’s o1 moves to ‘level 2’, says Sam Altman
OpenAI CEO Sam Altman, speaking at T-Mobile Capital Markets Day 2024, earmarked an incoming steep curve of AI improvement due to the o1 model.
One of the people responsible for ChatGPT mentioned that the o1 model is at the GPT-2 stage of reasoning and has been vocal about the exponential difference between the o1 preview release and the upcoming final release.
Earlier this year, Altman stated that the current version of the company’s AI is the “dumbest” at a seminar at Stamford University and would be surpassed by future builds by OpenAI.
In a recent interview with T-Mobile, Sam Altman compared o1’s current state to the ‘GPT-2 stage’ of reasoning models
He also revealed that the development of o1 unlocks a much quicker path to fully capable AI agents
Hear it from the man himself:pic.twitter.com/XeXFDstbT2
— R…
— Jay • 24Yr Old Millionaire (@thejacobdean_) September 20, 2024
Sam Altman has high hopes for o1 to create a generational leap in AI models
The o1 is centered around advanced reasoning for problem-solving, which Altman references heavily in the interview. Altman says that the experience is one in which users will not want to go back but only forward when it comes to the new models, which will eventually lead to a GPT-4 equivalent and then outpace the existing models.
Altman outlined five levels of AI development, with o1 at level 2 (reasoners), and implied that level 3 (agents) could follow in a short space of time. As part of the event, Altman stated “in the coming months, you’ll see it get a lot better as we move from o1-preview to o1, which we shared some metrics for in our launch blog post.”
The o1 launch post breaks down the forecasting that OpenAI sees as the next big generational leap for learning models. The post also highlights the lighter version of the o1 preview, the OpenAI o1-mini, which is also 80% cheaper than the preview.
As part of a research process of o1, both the UK and U.S. AI security and safety regulators have been involved in developing the new model, and OpenAI has produced a security framework.
According to the company, the framework is a living document intended to “process to track, evaluate, forecast, and protect against catastrophic risks posed by increasingly powerful models.” Altman, as part of the T-Mobile event, reiterated that the model and the development of its capability are not trained on sensitive data at any point in time.
Altman is confident that the problem-solving AI will be adopted by users quickly, but did reflect on the struggles that people had getting to grips with using previous OpenAI products. He said, “ It took people a while to figure out how to use ChatGPT, and it took us a while to build. It’ll take users a while to figure out how to use it, which is quite different from the GPT models.”
Image: OpenAI.
The post OpenAI’s o1 moves to ‘level 2’, says Sam Altman appeared first on ReadWrite.