On Monday, August 10, 2026, at 7:08 PM, Matt Mahoney wrote: > Of course he was wrong. We now have LLMs that pass the Turing test, which > means they have all the properties that he argued were not computable. It > turns out you don't have to model microtubules to build a neural network, > just like you don't have to model quantum tunneling of gate electrons in SSD > memories to model the digital logic of a computer.
The Turning test was low hanging fruit on the AGI tree. People would rather transmit in their language with the an AI peer that understands them as a human verses using some translation layer. That’s just simulating simulations. If I say I have a headache to another human they kind of get it, a sharp headache, dull, pounding, just transmitting approximations. I could transmit 1000 pages of syntactically unambiguous language and lack precision. How many bits would it take to transmit a full description of my headache? Still an approximation. My headache might be or actually is unique. You'd have to snapshot all the microtubules which I don't think is possible so an AI can only estimate especially when using only fuzzy words. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T906c878e4ba411be-M6d877187407507cab11ac66b Delivery options: https://agi.topicbox.com/groups/agi/subscription
