AI is completion generation. if appropriately constrained, prompted, and monitored it can generate completions at par with or even exceeding the best human material. the question is is the control cheaper than a human expert given it requires a human expert to monitor anyway, if the controlling an llm to get valuable output is cheaper than just relying on human experts which arguably it is who captures that value the user or the "ai platform" it seems to me clearly the user in the long run. nobody has ever sucessfully monopolized reasoning. where the myth starts is with the hypothesis a firm is going to discover agi <which is itself a term showing lack of understanding: intelligence is not generalizable but always applied>. Nature (physics) does not admit a duality with artificial. everything physical including ai is natural. the punchline is the notion humans can construct a superior intelligence and "own" it is nonsensical. There is no moat possible. Therefore it is a bubble.

then you see people like karpathy out there telling people to just vibe it out, ignoring the fact that he has like two decades of experience to wrap his head around failed/correct output