Interesting! Then what are your thoughts on papers such as the following that do ground large language models with interactions with the world?

I'm specifically curious about your thoughts on Zero-shot Common Sense and LM-Nav, since I don't think either are really grounded in the way you want, but they at least both appear to be a "truly intelligent robot moving around and performing [in this case, simple] tasks"

1. Extracting Zero-shot Common Sense from Large Language Models for Robot 3D Scene Understanding:

https://arxiv.org/abs/2206.04585

2. LM-Nav:

https://sites.google.com/view/lmnav

(Optional:)

3. Generally capable agents emerge from open-ended play:

https://www.deepmind.com/blog/generally-capable-agents-emerge-from-open-ended-play

4. Building interactive agents in video game worlds: https://www.deepmind.com/blog/building-interactive-agents-in-video-game-worlds