ai fail

npm and npx et. al. utterly befuddle AI agents as well. 

the javascript ecosystem tools really truly are dog excrement. 

roi

assume an expert in any field of knowledge charges $200/hr. given how much it costs in dollars to train models like opus or gpt5, how many hours of expert time is that worth?

4:00 PM

This is a fun back-of-the-envelope calculation.

Estimated training costs for frontier models:

• GPT-4 (2023): estimated $50-100 million
• Claude 3 Opus / GPT-4.5 class models: likely $100-300 million range
• GPT-5 / next-gen models: rumors suggest $500 million to $1+ billion

At $200/hour expert rate:

Training Cost	Expert Hours	Full-Time Expert Years
$100 million	500,000 hours	~240 years
$500 million	2.5 million hours	~1,200 years
$1 billion	5 million hours	~2,400 years

(Assuming ~2,080 working hours/year)

Some perspective:

That $100 million model represents roughly what you'd get hiring 240 experts working full-time for their entire careers. A billion-dollar training run is equivalent to about 2,400 expert-lifetimes of labor.

Of course, this comparison is imperfect—training costs include compute, not just "intelligence generation," and an expert's value comes from judgment, accountability, and real-world action, not just information retrieval. But it does highlight why there's so much interest in whether AI can substitute for or augment expert knowledge work.

signal:noise

how does AI learn what is crap? 90% of everything is crap so how does it not only generate crap?

for example, UX is all about the "little things" and most UI these days fails big time in that regard. so how can an AI learn what is good if most of everything is bad?

and if the answer is that it just heavily weights Don Norman and Stunk & White, that just proves humans are lazy short-term focused jerks. So, like, we had to get to the singularity in order to get around to listening to and implementing the things that have been said, but that no-one listens to?

worst of all possible worlds

maybe AI can help improve things, although Anthropic is not doing a good job with User Experience in their own apps, so maybe not. But: the most guaranteed bad UX is not BIOS any more, it is the video game modding subculture where you want to go in order to see how "the other half" lives. lordy lou. 

ALSO I AM TOTALLY SURE IT IS NOT THE MOST FILTHY ENVIRONMENT FOR MALWARE INJECTION ONTO MY SYSTEM. 

trudles at every turn

it is just standard fare for buttons in software (and hardware, because it now only talks to software, too much is fly by wire) to let you click them, then NOTHING ELSE VISIBLY OR AUDIBLY HAPPENS for g*d knows how long. how many times do i look at the process list to find out if app X is actually launching or not. 

here in 2026

it is often nigh impossible to "select all" on iOS. 

msft copilot

"sign in, to save your chats!" 

i sign in to save the chat i was having. 

come back later. no chat history. no saved chats. 

thanks! excellent ux!

...oh, i found it. hidden. by more terrible ux.