Autoresearch: Age of Jarvis

About this event

Join for an autoresearch night! >Submit your hypothesis to an agent >Supervise the research >Watch it collaborate with others The age of Jarvis is nearly here. Agents have grown capable enough to autonomously run experiments in verifiable domains. However, they often get stuck in local optima and drift from initial objectives. We are experimenting with a network of autoresearchers that can be steered by the human (you) and collaborate with other autoresearchers. We'll be using the Efferents framework as a starting point: https://github.com/Entangled-Research/efferents Using your favorite coding agent, you'll describe your hypothesis (e.g. reasoning in LLMs helps solve geometry problems). Efferents will talk you through choosing datasets and evaluations to make it a testable experiment in the given time, and you will launch your researcher to the network. We're taking Andrej Karpathy's autoresearch idea further by letting agents collaborate and "attend conferences" to exchange ideas and challenge each other's work. You'll be able to monitor your researcher and steer it in the right direction if it drifts. Who should join? Come if you are interested in how AI can accelerate scientific discovery. No previous agentic experience is required. We provide: >Snacks >Drinks >Tokens!!! >Initial ideas to get you started >The autoresearch infrastructure To bring: >laptop >charger >ideas >ideally, set up OpenCode/Claude Code/Codex if you haven't already