Oh wow, you really cracked the code. So we designed the architecture, wrote the training code, picked the datasets, and tuned the hyperparameters, but somehow we have no idea how or why they reach conclusions. That's like building a car and claiming you don't know why it moves when you press the gas. They are not magic mushrooms that we grow either. They are mathematical functions optimized through gradient descent. Every layer and activation function was a deliberate design choice. The control argument is just as weak. We don't fully understand how nuclear reactors work at the quantum level either, yet we still build them with safety mechanisms just fine. And burning our planet for a select few? Give me a break. AI is used in disease research, energy grid optimization, improving weather forecasting, helping disabled people communicate. But sure, let's focus on your Skynet fantasy.

Yes AI models can be threats. Not because they are unknowable eldritch horrors, but because people misuse them, data privacy is messy, and bias gets baked in. Those are real problems, but your we have no idea how they work so they will kill us all take is just the intelligent design equivalent for tech bros.