Why in the News?

OpenAI has introduced OpenAI o1, the first in a series of advanced AI models under its ‘Project Strawberry’ initiative.
- This new model is designed for tackling more complex tasks in science, coding, and maths.

About OpenAI o1

This model has been built to approach problems like humans, carefully considering various angles before arriving at an answer.
It improves its performance by learning from different perspectives and checking its output for errors.
In trials, the upcoming version of the o1 model performed on par with PhD students in areas like physics, chemistry, and biology, and excelled particularly in maths and coding.
For instance, it solved 83% of problems in a math contest, compared to earlier versions which solved just 13%.
- In coding, the model ranked higher than 89% of participants.

OpenAI is also releasing OpenAI o1-Mini, an economical version designed for developers, offering similar reasoning capabilities at 80% lower cost compared to the o1-preview version.
The o1 model excels in generating and debugging complex code and is expected to assist in software development, data analysis, and problem-solving tasks.

OpenAI has introduced new training methods to ensure the safety of these models, improving their ability to follow safety guidelines and prevent AI jail-breaking.
- Jailbreaking is a form of hacking that aims to bypass an AI model’s ethical safeguards and elicit prohibited information.
In safety tests, the new version scored 84/100, a significant improvement from the previous 22/100 score.
The company is collaborating with UK and US governments on AI safety and conducting red teaming to identify and address any weaknesses.