OpenAI released on Thursday a preview of its latest reasoning model series, "o1," which has been trained to solve more complex tasks and take on more difficult problems.
The o1 model preview is now available for ChatGPT Plus and Team users. The tech giant cautions users to expect regular updates, improvements, and evaluations.
In a news release, OpenAI shares more about how they trained the o1 reasoning models:
“We trained these models to spend more time thinking through problems before they respond, much like a person would. Through training, they learn to refine their thinking process, try different strategies, and recognize their mistakes.”
We're releasing a preview of OpenAI o1—a new series of AI models designed to spend more time thinking before they respond.
— OpenAI (@OpenAI) September 12, 2024
These models can reason through complex tasks and solve harder problems than previous models in science, coding, and math. https://t.co/peKzzKX1bu
Furthermore, the company emphasized that people shouldn’t expect the latest series to be useful in the more common tasks assigned to ChatGPT, like searching the web for answers.
The o1 models are specifically geared toward solving harder and more complex problems in math, science, and coding.
For instance, developers from different fields can leverage o1 to design and run multi-step processes, healthcare researchers can use it to label cell-sequencing data, and physicists can utilize it to create complex mathematical equations for quantum optics.
Understanding how AI works is difficult for many, but people don’t need to really know its inner workings to make use of it.
OpenAI is taking the initiative to help people understand how they build their AI models, as well as their numerous uses through short videos.
Using simple video production techniques, the AI giant is able to reach out to its users and give them a sliver of understanding about how the technology works.
The Breakthroughs That o1 Can Bring
On top of an explainer video that discusses how the o1 was built, OpenAI released a series of videos showing how the reasoning model can be applied in different fields.
Catherine Brownstein, a geneticist at Boston’s Children’s Hospital and scientific director at Manton Center for Orphan Disease Research, demonstrates how OpenAI o1 helps her with her tasks.
“It’s impossible to be an expert in every single gene. There’s 20,000 genes; you can’t keep everything straight, but AI can,” Brownstein said.
“I go down a lot of rabbit holes that do not yield anything useful. And being able to increase the percentage of rabbit-hole to useful information is killer,” she added.
OpenAI o1-mini, a faster and more affordable model that’s also focused on tasks related to STEM and “reasoning without broad world knowledge,” was also released.
It is now available to tier 5 API users at just 20% of the cost of the o1-preview, making it a great option for ChatGPT Plus, Team, Enterprise, and Edu users.
o1-mini (70% correct answers) performs better than o1-preview (44.6%) and doesn’t lag too far behind o1 (74.4%) when it comes to solving math problems. However, it is outperformed by the latter when it comes to non-STEM factual knowledge.
What OpenAI released is just an early version of its reasoning models within ChatGPT and its APIs.
Alongside updates to these models, the company plans to introduce more features in the future, such as browsing capabilities, file and image uploads, and additional enhancements aimed at widening o1’s utility.
AI-generated content has seen mixed reviews recently.
Motorola’s “Style with Moto” ad done using OpenAI’s Sora has received positive reviews while its AI-generated brand film for Toys 'R' Us faced backlash.
