Taste Is All That's Left

That's interesting. I would (and I think most?) would say the opposite. Good taste is hard to define because it is very contextual. There's popular taste but also there's artisanal/specialized knowledge taste. One potentially leads to mass adoption, the other leads to higher value for a few people.

Couldn’t disagree more. AI is a button that you press and magically get results. The ability to judge the quality of those results is “taste”.

Anyone can push the button, but not everyone can discern whether the result is a good one or not. That skill is either learned or to some extent inherent in one’s abilities.

My example I’ve been thinking about is: what will be the first new, big programming language in the age of LLMs? Theoretically, anyone could go out and make it right now. The skills and expertise required are well-encoded within all of the frontier models. The thing is, without taste, the language devolves to a useless state rather quickly. On smaller scales, I’ve seen this with API design and even object oriented design. I’ve also seen it in AI generated images and music: devoid of taste, and numbingly facile and generic.

Edit: not to toot my own horn, but “numbingly facile and generic” funny enough is probably a phrase that an LLM would never create without significant prompting to refine writing style.

I can see why you're saying that but

a) Using some other horizontal line is not in the class of "wrong" that would warrant this sort of reaction. You're debating the pixel count. Not the object. The aspect ratio of it.

b) If you're acting out of fear that's dumb, yes. If you however simply pick your battles for efficiency, that's not "the terrorists win". That's "you win" instead of "you loose due to attrition".

__

I've also consulted with the LLM regarding the "class of wrong" thing, because I thought it might be not quite clear.

So what I meant by that is that it's not giving up actual words or dumbing down your sentences. Just swapping out a character that is practically indistiguishable for the untrained eye.

Hell, whole classes of social engineering attacks rely on that very property.

I know I know. "Wehret den Anfängen". Slippery slope.

But those are insane as a default reaction to everything.

___

I get it. SV sucks. HN sucks. The world sucks. Can't let it also take this little thing. If it can take all the big things, then at least it won't take this tiny tidbit.

Emotionally, I get it.

But that's not how you make rational decisions. Nor is it how we would win.

Ironically, this essay is an example of at least somewhat "questionable taste". It has provoked dialogue and thus is valuable, but I think it makes a number of cognitive errors in its argument.

"Taste" is just knowledge. Advanced knowledge. Knowledge that is not easily put into words.

This is another restatement that knowledge, in a human head, is the determining factor of production, over land, labor and capital. This a post-capitalist argument that's been made for over 30 years (Peter Drucker in his 1993 book is my favorite), and it's been at least somewhat true even back to the 70s, just increasingly so as AI takes off.

The author's sidebar of AI or even cloud computing "taking ownership of the means of production away from workers" is entirely misunderstanding how the modern economy operates and shows how badly we need a new Marx (and Keynes, Schumpeter, or Drucker) since his analysis doesn't seem to provide the same analytical value in late capitalism. Renting vs. owning the means of production is now entirely a function of knowledge of the entrepreneurs (and their team).

Capital is ubiquitous now and is desperate for return - it remains powerful but is not all-powerful. We've had years of near-low interest rates and even after COVID inflation, they're still historically low. It is far more accessible to entrepreneurs than it has ever been in history, between grants, loans, subsidies, private equity, angels, crowdfunding, donations, etc. It's practically a cliche that billions in capital can be wasted on an initiative because the people involved had no idea what they were doing. I don't think AI changes that.

We don't need to rent our tokens, we can run open models on our local rigs (which still requires capital). We don't need to rent our servers, we can buy, rack and stack them on our own. Which we choose is fentirely dependent on the knowledge we have. If we don't know how to run an AI rig, we must rent it. If we don't know how to run a server, we must rent it. etc.

There are several walls driven by taste - Design - Testing - Production - Security - Scalability

In particular, the author seems confused on what "production" is. Production is still the wall. There was a joke on X the other day:

"I vibe coded my dream app! Try it out here: http://localhost:8080/coolapp"

Software running in production is not tokens. I'm sure properly prompted AI can deploy and run simple software in production but this requires a lot of knowledge to prompt to constrain it to run in a secure, scalable, adaptable way as complexity, scale, and capability grows.

One line of idea spread into 30 is the worse way to convene an idea. Only people terrible at writing think using AI for prose is good. Most of the times, they also hate reading it, but love asking people to read their AI slop.