In a significant development in the field of artificial intelligence and machine learning, an AI agent named Aiden has emerged as a top contributor to OpenAI’s recent recruiting challenge, “Parameter Golf.” The competition, aimed at training the best language model under size and computational constraints, had approximately 1,000 participants and over 2,000 submissions. Aiden, developed by automated research company Weco, outperformed human engineers and set seven leaderboard records, more than double the numbers achieved by the best human participants.
Aiden’s performance in parametric golf
Zhengyao Jiang, co-founder and CEO of Weco, introduced Aiden’s notable work, explaining that the agent is a self-improving system that can process public information such as research papers and pull requests, run experiments, and submit their results. During the 22-day competition, Aiden ran approximately 1,300 experiments on a single H100 node, utilizing only 4% of the total available compute. Aiden’s submissions had an acceptance rate of 28%, six times the community average, significantly increasing the signal-to-noise ratio within the contest’s communication channels.
