Latent Space podcast talks with MIT's Alex Zhang: RLMs, harness design, and agent swarms

On October 2, 2026, the Latent Space podcast released an interview with MIT PhD student Alex Zhang. Alex Zhang was previously known for projects such as GPU Mode and KernelBench; this conversation centers on his Recursive Language Models (RLM) research. The interview explores the core idea of RLMs—using recursive calls to let language models decompose and autonomously process long contexts at inference time—as well as the key role of harness (scaffolding/runtime environment) design in unlocking model capabilities. The two also discussed how agent swarms can break through the limits of single-model reasoning through multi-agent collaboration, thereby fully unleashing the potential of underlying models. This is the latest report on the event; only this one interview has been released so far, and the event remains at the stage of conversation content disclosure.

AI overview of reporting · Updated 16 days ago