Comment by ecshafer
8 hours ago
I regularly find the code output of opus and gpt 5.5 to be garbage. Overly verbose, unnecessary abstractions, strange duplication of concepts across objects, unnecessary copying of objects and creation of objects. I have found its much more useful to just ping pong some ideas, have it generate helper methods, and do the code implementation by hand.
I guess I am a stoneager.
I think this is one of the grand realisations, and it is why local models are much, much more of a threat to the "metered intelligence as a global utility" business model.
As an AI-cynic I am much more interested in learning how AI solves my problems (of which I have many), not how it can revolutionise programming. How about it revolutionises me not experiencing task paralysis first.