LLMs Are Not AGI

By Leo Gura - September 12, 2026

AI companies have started to watermark their own AI output text. Why?

Not for the reasons you’d think, not so humans can identify deep fakes, but because these companies have a serious practical need to differentiate AI text from human text, because if the AI models train on AI text, they degrade and turn to shit.

If AI was truly intelligent, then it would not turn to shit from training on its own text. But these LLMs have a huge problem when trained on their own outputs. This demonstrates how limited these LLMs are and how parasitical they are on human intelligence and creativity.

If AI is so much better than humans, then why do all these AI companies steal every piece of human-generated data they can find? Surely the AI is intelligent enough to not need it. Or not.

If AI companies have built AGI, then why are they still stealing and scanning books? Why would you need obscure human books if you had an intelligence greater than human? Why are humans not busy reading every book on the planet? Because that isn’t needed to be intelligent. In fact, memorizing every book on the planet does not make you intelligent, but allows you to fool many people into believing your are intelligent. Memorizing Wikipedia is not intelligence, but there is a lot of intelligence baked into Wikipedia.

Click Here to see ALL of Leo's juicy insights.