@RokoMijic imo 90% right- you do need interaction w the real world... but LLMs w better prompting & synthetic data & agent wrappers that interact w the world can slowly elicit the needed grounding to continue learning ... but at the world's pace not at gpu speed