If you thought artificial intelligence had peaked, get ready for the next step OpenAI has introduced its new o1 model, which pushes the boundaries of what we thought was possible.
But don’t worry, it’s not the beginning of robot domination-at least not yet.
What’s so special about o1?
O1 is not just another smart chatbot.
It’s a model that actually“thinks” before answering.
Yes, you heard right.
It goes through a long internal “chain of thought” before giving you an answer.
Think of it as an AI that sits down, thinks and carefully considers all the options before it speaks.
More info on the OpenAI website
This new approach is yielding unexpected results.
O1 is ranked 89th.
percentile in the Codeforces programming competitions.
This means it beat most human programmers in a competition where they are already solving quite complex problems.
For example, in the American Mathematical Olympiad AIME, o1 would rank among the top 500 students.
Yes, among those high school students who solve complex math problems with ease.
And on science tests in physics, biology, and chemistry, he outperformed even human experts with PhDs.
How does it work?
O1 was trained using reinforcement learning, which allows him to effectively use the “chain of thought” technique.
He learns to improve his thinking, correct errors and try different problem solving strategies.
It’s as if he has his own thought process in his head that allows him to solve complex problems.
Imagine you need to solve a complex math problem or program a complex algorithm.
O1 is able to decompose the problem into smaller parts, analyze them and find the optimal solution.
As a result, o1 can write, for example, efficient code to solve difficult problems that would take human programmers hours in seconds.
The only disadvantage of the model is that it is still only in a test “preview” mode and the price of tokens via API is much more expensive than, for example, the recently introduced smaller model based on GPT-4o.
But it’s logical, the model works with your input repeatedly and much more than the “basic” ChatGPT.
OpenAI also introduced o1-mini, a smaller and faster version of the o1 model that is optimized for STEM (science, technology, engineering and math) tasks.
The o1-mini is 80% cheaper than the o1-preview and yet achieves similar results in math and programming.
