No AGI without neurosymbolic AI - Gary Marcus
A bit of a critique about the current LLM hype - Gary argues that scaling alone (more data, more compute etc) will not get us to AGI, and that current systems are unreliable.
He offers his own suggestions on what is missing, but it's nice [comforting for me at least] to see that the current "state of the art" AI with all the money and resources thrown at it is still a bit shit (obviously some very technologically impressive shit in many cases, but far from reliable or "general" intelligence).