Multi-Stream LLMs: new paper on parallelizing/separating prompts, thinking, I/O

Researchers propose Multi-Stream LLMs, a new architecture that enables language models to simultaneously process multiple parallel streams of input, thinking, and output instead of operating in a single sequential stream. This approach allows AI agents to read while generating output, think while acting, and perform other simultaneous tasks, which addresses fundamental limitations of current chat-based models and improves efficiency, security, and monitorability.