Notice many of the posts on the front page are tools to use ai, not end products or actual value creating things. What if the real power of llms is in its ability convice humans of the technology's own importance leading to the present economic dislocation What if 'AI' is the mimetic virus of human cognition swallowing its own tail, believing its own Mythos? The reality is all sytems in nature are feedback constrained by their substrate and there is no new physics. So when innovation stagnated to perpetuate the myth of growth we made the myth of generative AI as a growth engine which it can not be.
I think you’re onto something here. I saw a YouTube clip referring to the Gel Mann Amnesia effect, and how when you hear people say things like “AI is terrible at front end by great at backend” in many cases that’s because the speaker knows front end so they can see the mistakes the LLM is making. The speaker doesn’t know backend so when an LLM produces code that “works” the speaker is blown away.
Speaking for myself, it’s the other way around. I see its weaknesses in backend work but am impressed (initially) by its front end abilities. Only after wrestling with it for a month and realizing that a good front end Eng could’ve made me a better and more reliable app in a week does the illusion start to fade.
Given the enormous amounts of money invested in this I don’t know how long the myth can sustain itself. For the record I’m not saying AI can’t be useful and can’t accelerate your work but it cannot replace jobs and I have yet to see a ten person startup vibe code a competing product with a leading SaaS company.
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
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.