@hi_cial I wish I could boost this one thousand times.

Even if LLMs and text-to-whatever models were ethically squeaky clean (in terms of training data, environmental impact, the whole nine yards), always produced the right answer, and were better at their jobs than humans, I still would not use them, for the same reason that I do not copy-paste from StackOverflow and the same reason that artists discourage tracing. I like actually building skills, and LLMs do not allow me to do that.

The fact that LLMs consistently produce less efficient, more vulnerable code than humans while leaving whole communities without reliable electricity and water to power their datacenters, and that their training data was created by people who were not compensated at all and labeled so that it could be trained on by people paid starvation wages in conditions described by workers as "modern-day slavery", is just icing on the cake.