If I understood correctly, it seems like OpenAI's leader is implying that large language models (LLMs) have something significant to contribute to robotics. This strikes me as curious because, as far as I know, these are two very different problem domains. Only one (LLMs) has experienced rapid advancements recently, while the other (robotics) hasn’t.

Robotics in the physical world demands fast and precise responses. A robot walking to a grocery store and picking up items, for example, needs quick and accurate decision-making to avoid falling, breaking things, or picking up the wrong items. LLMs, on the other hand, provide slow, approximate answers that often resemble truth but can include hallucinations. This works well for paraphrasing non-exact information but seems ill-suited to the precision robotics requires.

So, my questions are:

1. Have LLMs contributed to advancements in robotics? 2. Is there any reason to believe they will?

From my understanding, the current paradigm of LLMs relies on scaling compute power for diminishing returns, which is the opposite of what useful robotics needs: efficient and accurate computation.