Tools may change, but taste is hard to fake.

Michael Flarup in The Future is Colourful and Dimensional, via Daring Fireball, on generative AI and icon design:

Tools may change, but taste is hard to fake.

The type foundry Mass-Driver in From human hands., on why taste needs a human behind it (and citing the same left-handed writing system as Sinistral):

When we ask Midjourney to draw an ox, it does not think of home. ChatGPT will never invent a new calligraphy technique, because it has no wrists to feel uncomfortable. A tool that shields us from the friction of the work is compelling, but if we don’t experience the friction, we will never change the work.

and

on what gets lost when the tools set the taste:

If we allow these tools to dictate our visual culture, we must accept that it will be vitrified by machines which consider a few billion webpages equivalent to the sum total of human existence. I don’t accept that.

and

where the easy road goes:

AI might be a scenic path with no hills or hindrances, but it does not lead to the goal we’re chasing. It leads to a fucking desert.

The same goes for words. Here's Adam Mastroianni in 28 Slightly Rude Notes on Writing:

The beauty ain’t in the necklace. It’s in the neck. … Maybe that’s my problem with AI-generated prose: it’s all necklace, no neck.

and

what the neck is:

it’s not about filling in the biographical details, it’s about realizing that the thing you’re reading comes from somewhere. Good writing is thick with that coming-from-ness.

Same goes for code, even code an agent writes for you. See Robin Sloan's third rule in 'Ask for what you want' is a key skill for the 21st century:

Don’t let the AI agent name the app. You pick the name.

More on Robin Sloan's and Maggie Appleton's takes on what these tools can and do flatten are in A.I. Good vs. Bad.

Paths from here

• Give a crap: taste as a kind of care
• A.I. Good vs. Bad: what the tools can and can't fake
• Bespoke Software: new tools, same judgment
• My Personal A.I. Policy: rule 6, my voice in whatever I make
• Sinistral: a hand the training data never had