Brief IA
›

At Caltech, Mathematicians Build Their Own AI

Models & LLM·Tom Levy·

At Caltech, Mathematicians Build Their Own AI

At Caltech, Mathematicians Build Their Own AI
Key Takeaways
1At Caltech, Sergei Gukov and a few students are developing their own open-source mathematical models.
2Most of these models are limited to a specific problem; a more general tool is in development.
3Presented on Friday by Terence Tao, Open Math OS allows for data usage control.
Why it matters : the mathematical community is divided on the use of AI and criticizes OpenAI's practices.

At Caltech, Professor Sergei Gukov is developing open-source AI models designed specifically for mathematics, alongside a few students. This initiative is drawing attention from a divided community regarding the solutions provided by OpenAI.

Open Math OS Gives Control Over Data

Two mathematicians have accused OpenAI of using their conversation histories to help solve the Navier-Stokes problem. Control over data is precisely one of the features offered by Open Math OS, a platform presented on Friday at Caltech by Terence Tao, a mathematician from UCLA.

The nonprofit organization founded by Tao created this tool, which combines advanced open-source AI models, cloud computing resources, and collaboration functions. Mathematicians can decide how their data is used. Tao has been advocating for the development of AI systems by mathematicians themselves in several talks.

Models Remain Limited to Specific Problems

In the lab he leads at Caltech, Sergei Gukov is working on systems that mostly tackle a specific problem: asking them to switch to another domain causes them to fail. In contrast, tools from Anthropic respond to requests across multiple mathematical disciplines. Gukov indicates that a more versatile model is under development, but it is not ready yet.

During a sushi meal at a Japanese restaurant in Pasadena, the professor detailed the creation of his models with his students, starting from scratch. He likens them to Ferraris: high-performance machines that are difficult for the public to access. He also claims that some problems were solved with his tools before those of OpenAI. According to him, practicing with these systems enhances understanding and then mastery.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Born in Moscow, Gukov moved to the United States to study mathematics. He views this research as an opportunity to advance the discipline rather than as an existential crisis. To express his enthusiasm, he invokes the exploration of unknown territories by Christopher Columbus.

Mathematicians Challenge OpenAI's Practices

In October, OpenAI provided solutions to hundreds of problems deemed difficult, which leading mathematicians had struggled with for years. Gukov claims to have anticipated the danger. Major AI companies quickly disseminate their answers without prior review by academic specialists. On Substack and in lecture halls across the country, mathematicians are denouncing these practices, accusing OpenAI in particular of instrumentalizing their discipline to maintain enthusiasm around an IPO.

Many of these critics find the answers poorly constructed, even incomprehensible without weeks of deciphering by a researcher. Tasmin Chu, a mathematics PhD student at Caltech and host of a popular anti-AI publication on Substack, sees the systems designed by mathematicians as an attempt to "own the means of production." However, she considers the topic complex and withholds judgment until she sees how these initiatives materialize.

Chu feels aligned with mathematicians who reject any use of AI in their work and advocates for the possibility of pursuing mathematics without it. Other students are fully engaged: some believe AI will save the discipline, while others think it does not deserve to be saved.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.