OpenAI Develops a New AI Model Capable of Reasoning and Solving Complex Problems

Think before you speak. It is a maxim with little traction among humans, but now it will govern a new generative artificial intelligence system. OpenAI, the company behind ChatGPT, unveiled yesterday the o1 series of models, aiming to make a qualitative leap in the sector. "They are designed to think longer before responding and can reason about complex topics to solve more difficult questions in science, programming, or mathematics," explains the American company.

The novelty of this definition lies in reasoning. "We have trained these models to think about the problems presented to them as a person would. Through this training, they learn to refine their thought process, try different strategies, and learn from their mistakes," OpenAI explains, claiming to have achieved results similar to those of doctoral students.

Their main capabilities are in programming and mathematics. The company provides a surprising example: if ChatGPT-4o managed to solve 13% of the problems posed in the qualifying exam for the International Mathematics Olympiad, o1 achieved an 83% success rate. "Their code-generating skills were evaluated in various competitions and ranked in the 89th percentile of participants," adds OpenAI.

This does not mean that o1 will replace ChatGPT. At least for now. "It is an initial version that lacks useful functions such as browsing the internet for information or uploading files and images. In many common use cases, ChatGPT-4o will continue to be more capable in the short term," the company states.

Nevertheless, o1 serves as a wake-up call regarding what AI can do. And it will surely give wings to those who warn about the dangers this technology poses to humanity. Aware of this, OpenAI emphasizes that the model "has been trained to always follow safety protocols and stay within set boundaries," and seeks to reassure the public by stating that it scored 84 out of 100 points in tests designed to make it break the rules, while the latest version of ChatGPT only scored 22. This means it is practically unbreakable.

For now, both the standard version of o1 and the 'mini' version—80% more economical and mainly intended for coding—will be available to users of ChatGPT's Plus and Team versions (Enterprise and Edu starting next week), with usage limited to 30 and 50 messages per week for o1 and o1 mini, respectively.

$150 Billion

The advances OpenAI is making in Artificial Intelligence are turning Sam Altman's startup into an extraordinarily valuable company. So much so that with the acquisition of another $6.5 billion in its latest funding round, still ongoing, its valuation could soar beyond $150 billion, according to Bloomberg estimates. This would make it one of the most valuable unlisted startups, alongside ByteDance—TikTok's parent company—and Elon Musk's SpaceX. In the latest round of fundraising, two of the world's leading tech companies are reportedly considering participating: Apple and chip manufacturer Nvidia.