@felipe @grumble209 An AI-head I read early on said a project centering LLM-written code suggested a need for (ballpark) 10:1 test code to production code, vs. 1:1 with human-written code.

Your numbers seems in line with that, and maybe the tests get good enough to try tape out.

"AI" could go like the fiber part of the dot-com bust (with differences): lots of overbought/built GPUs and data centers (instead of fiber) that do get used but not as soon & profitably as banked on.

@Rycaut Great real-world example of how complex & intertwingled even 'simple' lists can get! You're maintaining and working through, is it a progress list?, for the year. Others might maintain 4 lists:

Books I have read
Books I would like to read
Books I own
Books wishlist, shopping list

People want to organize as they like. (And then the dev inquiry is how much back end data can be available interoperably, while still having front ends that can match any given user's idiom, ui prefs, etc.?)

@Rycaut Defaults are so impactful, I fear a tech future where more & more if them are set to serve extractive profits. And more generally, a world there aren't enough people who know how to set them.

Or even just enough people who know how to change them!

It's bad enough now and LLMs — or even just the hype around them — make this an inflection point that could could make it worse, or better, as they go from cannibalizing search and junior developers, to whatever they focus on next.

@Rycaut I mean, I appreciate language translation, and really don't care if the back end is an LLM or some other machine/artificial intelligence tool. 🤷

LLMs have been so overhyped and overinvested in. :-P Ever since the image models and then OpenAI strarted to take off, I've had deep critiques that have gradually gotten clearer.

The $ bubble is many tragedies, just waiting to happen.

Wish there was more attention, and investment, in public interest efforts. https://publicai.network/