I think we're losing on calling LLM output slop. Because... that's not so universal as it was. A current model prompted by someone who isn't a complete idiot can produce useful results. They still need to be reviewed and tested, but the results of vibe coding can be fit for task. It won't be long before that's just a losing argument, period, at least for vibe coding(control software for safety crtiical systems is another matter) The ethical arguments though? The massive resource use, the sourcing and licensing of the training data? Those arguments still hold up, and without fundamental redesigns of the entire pipeline will be very difficult to solve to the extent they've solved the problem of useful output. And Open Source developers might get blinded by a tech shiny from time to time, but they can also be grabbed with a well reasoned argument on ethics around software. But they're not getting that argument, they're just getting somewhat outdated getting more outdated by the day arguments about resulting code quality.