OpenAI has introduced a new version of ChatGPT based on a new artificial intelligence technology called OpenAI o1. What are the new features, and why is the new version better for scientists?
On 12 September, OpenAI presented a new version of ChatGPT virtual assistant based on the new artificial intelligence technology OpenAI o1. Since the launch of the first version of ChatGPT in 2022, the chatbot has been constantly improving, providing users with new features. In this article, we will review the latest updates and find out how the new version differs from the previous ones and why it is better.
Demonstration and features of the new ChatGPT
OpenAI has announced that the chatbot based on OpenAI o1 technology is able to “reason” while solving maths, coding, and science-related tasks.
During a demonstration for The New York Times, the chatbot was presented with some puzzles and asked questions about chemistry, which it answered at the level of a PhD, and diagnosed the disease based on a detailed report of the patient's symptoms and medical history.
The company also noted that the new technology could help physicists generate complex mathematical formulas and assist healthcare researchers in their experiments.
Experts have trained these models to spend more time analysing problems before providing an answer, mimicking the human approach. Through this learning process, the models improve their thinking, try out different strategies, and are able to admit their mistakes.
The need to improve artificial intelligence
ChatGPT has been trained by analysing large amounts of text from various sources on the Internet, including Wikipedia articles, books, and chats. By analysing patterns in texts, it gained the ability to generate new text on its own. However, due to the prevalence of false information on the Internet, the model can reproduce these inaccuracies and sometimes even invent them.
The developers have created a new OpenAI system using a reinforcement learning method, whereby the system learns through numerous trials and errors, which can take weeks to months. For example, when solving mathematical problems, the system identifies which methods lead to the correct result and which do not. After completing numerous such tasks, it begins to notice patterns, but this does not mean that its thinking is similar to human thinking. OpenAI technicians emphasise that the system can still make mistakes and is not perfect, but users can expect it to work harder and be more likely to provide correct answers.
Testing the new OpenAI o1 technology
OpenAI said that the new technology performed better than previous technologies on some standardised tests.
In the tests, the new version of the model performs at the graduate level in challenging physics, chemistry, and biology tests. The model also shows excellent results in mathematics and programming. In the AIME 2024 exam, the GPT-4o model was able to solve only 12% (1.8/15) of the problems on average. At the same time, the o1 model achieved 74% (11.1/15) of solutions with a single approach to the problem, 83% (12.5/15) with consensus among 64 attempts, and 93% (13.9/15) when re-ranking 1000 attempts using the learned evaluation function. A score of 13.9 places a student in the top 500 nationally and exceeds the cut-off score for participation in the US Mathematical Olympiad.
The model, initialised on the basis of o1 and enhanced for programming, scored 213 points and was ranked in the 49th percentile at the International Computer Science Olympiad (ISPO) 2024. In the real-life competition, 10 hours were allocated for solving 6 algorithmic problems, with 50 attempts for each. Applications were evaluated based on public and generated tests. If the answers were chosen randomly, the average score would have been only 156 points, indicating that the strategy used added almost 60 points in the face of fierce competition. With the number of attempts relaxed, the model achieved a score of 362.14, exceeding the threshold for a gold medal. On the Codeforces platform, the GPT-4o model received an Elo3 score of 808, which is in the 11th percentile among humans.
Access to the new ChatGPT
Access to the new technology started on 12 September for consumers and companies that have subscribed to ChatGPT Plus and ChatGPT Teams. The company also sells the technology to software developers and companies that have created their own AI applications.
Users of ChatGPT Enterprise and Edu will have access to both models from 16 September. OpenAI plans to provide access to o1-mini to all free ChatGPT users, but the exact release date has not yet been determined.
Developer access to o1 is expensive: the o1-preview API charges $15 for 1 million incoming tokens and $60 for 1 million outgoing tokens. For comparison, using GPT-4o will cost $5 for 1 million incoming tokens and $15 for 1 million outgoing tokens.
New opportunities for researchers and scientists
The new OpenAI o1 technology opens up many new opportunities for researchers and scientists, including:
- Better analytical capabilities. Thanks to the rapid solution of complex problems, scientists will be able to analyse and find the right solution faster. The new technology also allows for faster processing of large volumes of data, which significantly saves time when conducting research.
- Optimisation of the literature review. The model can quickly process numerous scientific sources and help create high-quality literature reviews, which saves time for researchers when preparing articles and studies.
- Improving text quality. ChatGPT o1 will allow authors to edit the text of a scientific paper more efficiently and accurately, correct errors, and make suggestions for improving the quality of the text. This is especially important for those authors who are preparing a paper in a foreign language.
o1 – and its successors – will open up many new opportunities for using AI in science, coding, mathematics, and related fields. In the near future, users and developers will learn how this can improve their daily work.
The new technology from OpenAI makes the preparation of scientific research and writing of scientific papers more efficient. However, it should be remembered that it is not perfect and can make mistakes. While tools such as ChatGPT can be useful in preparing articles, they can never completely replace the personal input of the author.
Subscribe to our social media to keep up to date with the latest news and useful information.
