Lambda is a good name; the training method is also lambda—functionally generating understanding.

April 17th, Carnegie Mellon University's Robotics Institute and Lambda jointly proposed the Sim2Reason training method, aimed at addressing the scarcity of high-quality data in the STEM AI field. The core is to perform unsupervised training of large language models in a virtual world governed by real physical laws, allowing them to learn physics through experiential learning. The method claims to improve the model's zero-shot performance in the International Physics Olympiad by 5-10%. Source: InFoQ