Academia is for Ambition — Alex Zhang, MIT
MIT PhD Alex Zhang's Recursive Language Models first solved ARC-AGI-3 before OpenAI's Astra
“the “language model” of the future may actually be an invisible swarm of agents underneath a simple interface”
MIT PhD student Alex Zhang's research on Recursive Language Models (RLMs) produced the first system to approximately solve ARC-AGI-3, beating OpenAI's Astra, positioning agent harness design as a critical unlock for model capability. Zhang argues that current AI systems leave significant capability on the table by wrapping powerful models in primitive orchestration layers. The research points toward a future where 'language models' are actually invisible multi-agent swarms behind simple interfaces, with implications for how labs and researchers should think about system architecture.