As much as the remote AI stuff is good for, really. Local models have been catching up to commercial big-iron fare at an astounding pace. Last year I would’ve said they’re a year behind. But in April, Qwen released a 27B model that outperforms their 397B model… from February.

I’ve seen a variety of Youtube makers build a thing and then hand-wave some code with a nervous laugh. Even if it’s a coin toss whether the chatbot-that-codes actually works - you don’t lose much by trying. Epecially if the alternative was reluctant procrastination.