OpenAI Challenges Researchers with Its 16MB 'Parameter Golf'

OpenAI Launches an Ambitious Challenge for AI Researchers

OpenAI has recently announced a unique competition called "Parameter Golf," aimed at pushing the boundaries of language model compression. This challenge invites researchers to design the most efficient model possible while adhering to a strict size constraint of 16 MB for the total weight of the model and training code.

Participants must also comply with a time constraint, as model training must not exceed ten minutes and must be conducted on eight H100 GPUs. Submissions will be evaluated based on their ability to compress effectively against a predefined dataset, FineWeb.

One Million Dollars at Stake and Career Opportunities

To motivate participants, OpenAI, in partnership with Runpod, is offering one million dollars in computing credits. But the stakes don't stop there: top candidates may be offered job interviews. OpenAI plans to integrate a small group of junior researchers, including students and Olympiad winners, starting in June.

The dedicated GitHub repository for this competition provides reference models, evaluation scripts, and a public leaderboard to track participant performance. The competition is open to anyone aged 18 or older, residing in supported countries, and submissions are accepted until April 30.

Fierce Competition for AI Talent

The race for talent in the field of artificial intelligence is more intense than ever among tech giants. OpenAI is not the only company seeking to attract top researchers. Meta, for example, has already managed to recruit several experts from OpenAI, sometimes offering compensation packages that can reach 300 million dollars.