“AI centipede” is almost as great a term as “Habsburg AI”

I think it’s already happening, I read something about LLM providers finding a way to circumvent model collapse…

Available in TailwindCSS with the `starting:` variant

class="transition-opacity duration-300 starting:opacity-0"

.fade-in {
transition: opacity 300ms ease-out;
@starting-style {
opacity: 0;
}
}

I should probably alias focker=docker 🤭

So we managed to invent a new way for everyone to write features, and convinced everyone - including the people who already write features - that it doesn't count as real work.

AI isn't replacing labor so much as it's devaluing it.

And from what I'm seeing, a lot of developers treat prompt engineering like it's beneath them. No interest for testing, improving, evaluating, guardrailing prompts.

It's not "real" coding. Just wrangling words.

A year later: in most cases, developers are writing the prompts and hardcoding them in the source.

The product people who were going to be liberated? Haven't touched a prompt. Don't even seem to be interested in them.

So for the foreseeable future - which apparently nobody foresaw - I argued developers would be the ones writing and maintaining the prompts.

Also we've already been through this with copy text in web apps. Non-developers are supposed to write it, there are whole systems for managing it, and somehow it always ends up copy-pasted by developers from Jira tickets straight into the code.

At the time I argued that "prompts are carefully-worded text often with variable inputs that you send to a machine to interpret and produce results".

We have a word for that. It's called code.