OpenAI releases the "o1" new generation large model, which is better at reasoning and more expensive

OpenAI releases "o1", a new generation of large models, which is better at reasoning and more expensive

The legendary "Strawberry" appeared. On the evening of September 12, OpenAI officially released a new model called o1. This model is the first of the company's next-generation "reasoning" models. o stands for "Orion". This model can answer more complex questions faster than humans.

Compared with previous models, it is better at writing code and solving multi-step problems. But it is also more expensive than the previously released GPT-4o and answers questions slower. OpenAI emphasized that this release of o1 is a "preview version" and is only in its initial state. Also released at the same time is a smaller and cheaper version o1-mini. For OpenAI, o1 represents a step towards its broader goal of human-like artificial intelligence.

ChatGPT Plus and team users can access the o1 preview and o1-mini from now on, while enterprise and education users will get access early next week. OpenAI said it plans to make o1-mini accessible to all free users of ChatGPT, but has not yet determined a release date.

For developers, access to o1 is much more expensive than before: using the preview version of o1 through an API costs $15 per million tokens of input and $60 per million of output. In contrast, GPT-4o charges only $5 for a million tokens of input and $15 for output.

Jerry Tworek, head of research at OpenAI, told the media that o1 "is trained using a new optimization algorithm and a new training data set tailored for it," and it sets up a reward and punishment mechanism to train the model to solve problems on its own through reinforcement learning techniques. It uses a "thinking chain" similar to the way humans solve problems step by step. This new training method makes the model more accurate. "We noticed that this model has fewer hallucinations," Tworek said, but the problem still exists, "We can't say we have solved the hallucination problem."

According to OpenAI, the main difference between this new model and GPT-4o is that it can solve complex problems such as coding and mathematics better than its predecessor, while also explaining its reasoning process. OpenAI also tested o1 on the International Mathematical Olympiad Qualifying Exam, and while GPT-4o only solved 13% of the problems correctly, o1 scored 83%.

If you want to run OpenAI on your computer , dont forget to Buy Windows 11 and office 2021 at Keyingo.com

The emergence of the o1 model means that the reasoning ability of the large model can fully reach the expert level, which can be regarded as a milestone in artificial intelligence and will greatly improve the application of the model in the enterprise.

As the model's abilities in intellect, sensibility and rationality continue to improve, it will surpass human capabilities. It is difficult to predict what impact artificial intelligence will have on humans in the future. "The development speed of artificial intelligence now exceeds the speed of human cognition, and artificial intelligence governance will be a huge challenge.

The new model reached the 89th percentile of participants in online programming competitions known as Codeforces competitions, and OpenAI claims that the next update of this model will perform "similar to a PhD student" on challenging physics, chemistry, and biology benchmark tasks.

Currently, OpenAI uses human data to synthesize new data to enhance reasoning capabilities. However, synthetic data is limited by the original data and cannot synthesize infinite data or obtain essentially novel data. It cannot invent new disciplines or propose new theories like Einstein. "In terms of hardware, reasoning requires less computing power than training, but due to the extension of the thinking chain, the requirements for reasoning efficiency become higher, which puts higher requirements on the accelerated optimization of the reasoning process. However, with the improvement of large models in multiple capabilities, it has brought challenges to governance. The challenge is that the speed of human understanding of it is not as fast as its development speed.

Although it performs better in math and code, o1 is inferior to GPT-4o in many ways, including poor performance in factual knowledge about the world and no ability to browse the web or process files and images. However, OpenAI believes that it represents an entirely new category of ability, and it is named o1 to represent "resetting the counter back to 1."