The race to build smarter robots has a new dissenter. Agibot's chief scientist argues that robotics will not get its "GPT moment" by copying the approach behind large language models, according to KrASIA.
The comments, also reported by Let's Data Science, amount to a rejection of one of the most popular ideas in the field: that the same recipe that produced breakout chatbots can be scaled up to give machines general-purpose physical intelligence.
Why does this matter to anyone outside a robotics lab? Over the past few years, a dominant assumption has taken hold across the industry — that feeding ever-larger models ever-larger piles of data would eventually unlock a single, transformative leap, the way ChatGPT did for text. Many companies are pouring money into that bet. Agibot's chief scientist, per KrASIA, is publicly questioning whether the same path leads anywhere for robots that have to act in the messy physical world.
The sources here are brief, and they do not spell out the alternative approach Agibot favors or the technical reasoning behind the position. What they establish is the stance itself: a notable robotics company is charting a different course rather than following the LLM crowd.
That disagreement is worth watching. Language models work by predicting text, but a robot must perceive, balance, grip, and recover from mistakes in real time — challenges that may not yield to simply scaling up a text-style model. If Agibot is right, billions in industry investment could be flowing toward the wrong blueprint; if it is wrong, the company risks falling behind rivals who embrace the LLM path.
Why it matters: how the robotics industry answers this question will shape which machines actually make it out of the lab and into daily life.