Academia is for Ambition — Alex Zhang, MIT
MIT's Alex Zhang discusses Recursive Language Models, GPU kernels, and multi-agent harnesses.
Latent Space interviewed MIT PhD Alex Zhang about Recursive Language Models, KernelBench, and multi-agent harnesses. Zhang describes context offloading, code execution, and recursive subagents, noting an RLM-based harness approached ARC-AGI-3 before OpenAI's Astra. The discussion also covers AI-written GPU kernels, OpenAI's reported 10,000-agent experiment, Kimi swarms, and Sakana AI's open-ended research.
- Alex Zhang discusses RLMs that offload context and call recursive subagents.
- An RLM harness approached ARC-AGI-3 before OpenAI's Astra.
- AI-written GPU kernels still leave substantial room for human expertise.
- OpenAI's swarm trial reportedly used about 10,000 agents and 130 billion tokens.
- Products
- KernelBenchClaude CodeCodex
- AI models
- Recursive Language ModelsAstraKimi
From the original article
15 min read at latent.spaceWe catch up with RLM first author Alex Zhang, MIT PhD, on Jev, PhD masxing, and the future of harnesses.
latent.space
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.