OpenAI o1: a New AI Model That “Thinks Like a Human” Before Responding
- OpenAI unveiled a new AI model called OpenAI o1 that thinks like a human before answering. This allows it to solve more complex questions with better accuracy.
- The model is the first in the series – many more are expected to follow.
- Currently, it’s at a very early stage so it’s only available as a preview.
OpenAI unveiled a new AI model, called OpenAI o1 (codenamed Strawberry), which “thinks like a human” before answering a question. It has been designed to answer more complex questions from science, maths, and coding.
OpenAI also released OpenAI 01-mini, meant for faster and cost-effective coding and optimized for STEM reasoning.
These models are currently available to ChatGPT Plus and Team users in the preview mode. ChatGPT Enterprise and Edu users will get access to these tools next week. The preview is limited to 30 messages for o1-preview and 50 for o1-mini per week.
The new models are not ready for public release, but they’ll gradually be available to more ChatGPT users. OpenAI is also working on offering o1-mini to all ChatGPT users for free.
How Does It Work?
When you throw a question at OpenAI o1, it will analyze it like we do – from different perspectives and use different strategies to arrive at the solution. This refined process allows it to notice its own mistakes, increasing the accuracy of the final result.
- Giving an example of how it can be used, OpenAI said healthcare workers can use it to annotate cell sequencing data.
- Similarly, physicists can use it to generate “complicated mathematical formulas needed for quantum optics.”
While ChatGPT-4o solved only 13% of the questions in the International Mathematics Olympiad (IMO), the o1 model scored a whopping 83%.
The current version only thinks for a second before answering. However, according to Noam Brown, a research scientist at the company, they are trying to make the upcoming versions think for hours, days, or even weeks.
Limitations
There’s one obvious drawback to this approach – the energy bills will go through the roofs. But on the other hand, longer thinking time would mean the tool can solve more complex questions. Maybe it can help create a new cancer drug or find a cleaner energy source, who knows? The possibilities are endless.
The only question is what price we are willing to pay for these breakthroughs.
The tool has one major limitation. It can neither receive images nor search the web. This might narrow down its usage scope. But at the same time, we have to remember that it was intentionally designed for other purposes.
If you want to send image queries, you can always use ChatGPT. Whether these missing features will be added later or not is to be known.
Our Editorial Process
Related Post
The Golden Age of offbeat Arctic research
Enlarge / At the US Army’s Camp Century on the Greenland ice sheet, an Army [...]
Rail container dwells in Seattle-Tacoma still high, but relief in sight | Journal of Commerce
Dwell times at marine terminals in the Northwest Seaport Alliance have doubled in recent weeks [...]
1890 National Scholar’s Roots Run Deep in Agriculture
Posted by MD Sharman, Public Affairs Specialist, USDA Office of Partnerships and Public Engagement in [...]