Yi Ma

Training updates that improve performance in LLMs can actually degrade inference quality—unless you use the new Monotonic Inference Policy Update framework.

Achieving state-of-the-art performance with just 8 billion parameters, Embodied-R1.5 redefines the capabilities of embodied models in complex physical tasks.

Forget task-specific fine-tuning – teaching VLMs basic geometry yields a +29% boost on spatial reasoning benchmarks.

LLMs, like humans, exhibit a "frequency bias," performing better when prompted and fine-tuned with more common textual expressions.