eos

VoidPadding introduces a [VOID] token to handle padding in masked diffusion language models, allowing [EOS] to focus solely on semantic termination. This method significantly improves performance on reasoning and coding benchmarks while reducing decoding steps.

This paper presents an AI agent that integrates large language models with laboratory orchestration software, allowing scientists to create, monitor, and manage automated lab protocols using natural language. Evaluated on three simulated labs, the agent achieves a 97% first-attempt protocol generation success rate and requires far fewer interface actions.