New AI models from OpenAI are designed to think through problems more thoroughly
OpenAI is testing the new AI models o1-preview and o1-mini with advanced reasoning. They are also designed to solve complex problems in science and programming.
(Image: Hung Chung Chih / Shutterstock.com, Skulptur: Auguste Rodin, Der Denker)
OpenAI has released two new AI language models that promise enhanced reasoning capabilities. The o1-preview and o1-mini models are now available for the first user groups. The models have been trained using reinforcement learning to think through complex problems more thoroughly before providing an answer. According to OpenAI, this should make the models better able to cope with challenging tasks in science, programming and mathematics. As examples, OpenAI cites the annotation of cell sequencing data, the creation of complicated mathematical formulas for quantum optics and the creation and execution of multistep workflows for developers.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
In internal tests, o1-preview performed similarly to doctoral students in benchmark tasks in the natural sciences. In a qualifying test for the International Mathematical Olympiad (IMO), the model solved 83 percent of the tasks correctly, while GPT-4 only achieved 13 percent. In programming competitions on the Codeforces platform, it achieved the 89th percentile.
However, o1-preview still lacks some of ChatGPT's functions, such as the ability to search the internet or upload files. OpenAI admits that GPT-4 is therefore likely to be more powerful for many everyday tasks for the moment. In addition, the results with o1 depend very much on how long the AI is allowed to calculate.
In parallel with the development of the new models, OpenAI says it has also been working on improving AI security (PDF). This includes increased internal controls, cooperation with authorities and extensive tests and reviews.
In addition to o1-preview, OpenAI is also presenting the smaller o1-mini model, which is primarily optimized for programming tasks and is said to be 80 percent cheaper than o1-preview.
Very limited access
Access to the new models is initially restricted to ChatGPT Plus and Team users as well as API developers in "GPT Usage Tier 5". Usage is limited for now: Plus and Team customers can make 30 requests per week to o1-preview and 50 to o1-mini, while the aforementioned API developers can still make 20 requests per minute.
OpenAI intends to give ChatGPT Enterprise and Edu customers access to the new models from next week. OpenAI has also announced that it will later release o1-mini for free ChatGPT use.
(vza)
