Technology has never been perfect, but we've trusted it and its purveyors to fix the problems when they're noticed.
The problem with AI has been that a lot of people are trusting it unquestioningly, and when it spews out garbage, attitude has been "Meh, close enough." It'll get better, but only by double and triple validating the work, which means even more processing and power usage.
Those people who want to turn over full control over to LLMs are in for a rude awakening.