« the era of (auto) research »
the nanogpt speedrun is a great setup to test autonomous research: fixed model, one number to beat, and a human record that keeps moving. we pointed coding agents at it on idle compute and let them iterate for days, thousands of runs with minimal human intervention, until they beat the human baseline. in this talk we go through how they did it, how codex and claude code behave very differently as researchers, and why speedrun are one of the best environments we've found for studying autonomous research agents
About the Autoresearch Track
Autoresearch sessions at AI Engineer World's Fair 2026 in San Francisco.
When
Wednesday, July 1, 2026
12:05 PM - 12:25 PM·20m
Where
Main Stage
Capacity: 4000 attendees
Sign in to add this talk to your schedule.
Speaker
Elie Bakouch
Researcher
Prime Intellect
@eliebakouch
Researcher focused on training large language models; previously associated with Hugging Face and now working on open model training at Prime Intellect.