At the 2026 World Artificial Intelligence Conference, Galaxy General Robotics
founder and CTO Wang He said that with massive integrated multimodal data
training, embodied-intelligence foundation models could reach a 70–80% success
rate on tasks without task-specific training — comparable to early
conversational performance of language models. He identified powerful
pretraining combined with efficient downstream/post‑training paradigms as the
key enablers, and said that, given current data accumulation and model
convergence trends, the industry could see an embodied‑intelligence “ChatGPT
moment” by end‑2028.
