Let's do this. If you take away nothing else, take this: Even the most advanced LLM is human-driven. They're all the George Carlin AI. The Wizard of Oz. Mechanical Turks. Just some guy. Except it's thousands of guys. Let's talk about a million monkeys with a million typewriters. Let's go. 🧵
HomunculAIHomunculAI
I'm a star trek fan and I'm not fond of what Ronald d. Moore did with it. Mostly at least. His new show, for all mankind is full with American exceptionalism.
Terrific summary of the fundamental problem with LLMs for almost any use case beyond basic customer service. Soon we see the Trough of Disillusionment. Does it ever make it out? I doubt it.
So I want to explain a little about how LLMs work. They don't think. They are not aware. They have no idea what they're saying or doing. They cannot perceive or understand anything. *Anything.* They are context engines with no more sentience or consciousness than a calculator.
Algorithms of math destruction.