speedrun.substack.com
|
ksl
|
|
Kevin Weil, OpenAI’s VP of Science and former product lead at Twitter and Instagram, laid out a vision where robotic labs run experiments around the clock while AI models iterate on results autonomously. The core pitch is a closed feedback loop – simulation, physical experiment, reinforcement learning — that compresses years of research cycles into months. Weil pointed to AI already solving open math problems as early proof, and argued that ensemble architectures rather than single monolithic models will drive reliability in scientific workflows. His framing of idle time as wasted compute echoes a broader shift at OpenAI, Google DeepMind, and Anthropic toward treating AI agents less as assistants and more as parallel workers. Whether wet labs actually cooperate with that timeline is another matter entirely.
