Alex L Zhang

Recursive Language Models (RLMs) enable language models to process unbounded input context length by recursively decomposing and interacting with context through REPL environments, with GPT-5-mini outperforming GPT-5 on long-context benchmarks.

“Author of "Recursive Language Models" in Alex Zhang”

“Co-author on weight-space programming research with Takehi and Teoh.”

“Co-authored research showing frontier LLMs can directly read and write neural network weights to improve model capabilities.”

“Co-authored research on weight-space programming demonstrating LLM capabilities in weight manipulation.”

“Introduced Speculative Programmatic Tool Calling (sPTC) technique for overlapping tool calls with token generation, achieving 1.0-1.2x speedup.”

“Researcher who authored a technical post on speculative programmatic tool calling, a technique for overlapping tool call computation with code generation in language models.”

“Authored research article on language model harnesses as compositional generalizers, studying how recursive language models improve generalization across task lengths and domains.”

“Authored research thread on RLMs and compositional generalization, arguing harnesses enable task generalization.”

“Lead author of arXiv paper on Recursive Language Models that handle 10 million tokens with 58x better performance than standard models.”

“MIT researcher who authored paper on Recursive Language Models for processing long prompts beyond model context windows.”

“Reviewed drafts of the Chroma Context-1 research work.”