ShengShu livestreams two weeks of Motubrain 2.0 training before open-source release

ShengShu Technology has begun a roughly two-week livestream of the training process for Motubrain 2.0, the next version of its robot foundation model. The company plans to open-source the model soon after the broadcast, which covers pretraining, post-training, agent tasks and recursive self-improvement.

• The livestream began on October 21 and is set to run about two weeks
• Training is split into four stages: pretraining, post-training, agent tasks and self-improvement
• The first Motubrain launched in April as a world action model
• Model weights are not yet released and results are described as exploratory