Has someone made a timeline of all the times that the AI tech bros has shifted the goal posts on “AGI” and “superintelligence”? It must be a head-spinning list. And anyone who can read such a list without their head spinning probably has millions of dollars (or so they imagine) “at stake” in a fundamentally speculative technology.

Btw I’m fine of course if someone can point me to a legitimate counter narrative here, but Gary Marcus seems to be repeatedly elucidating some major functional hurdles that means the current tech either doesn’t or can’t easily implement the fundamentals of reasoning and context and fact-checking for discovering truth (not just consistency or frequency, because stronger pattern =\= more true, but in my experience LLMs absolutely rely on pattern inertia to “fill in blanks” and assume things more commonly stated are more likely to be true). His previous post about how Stanford researchers found LLMs hallucinating about X-ray images that they haven’t even seen yet is pretty shocking.