DK_en3x07 - LLMs Are The Air Guitar Of Intellectual Work

Episode first aired on 11 July 2025. Listen on Spreaker

First a correction: the phrase "Great is the confusion under heaven" is from Mao Tse Tung, not from the book of Ecclesiastes.
Heavenly points to James who noticed it.

Today's rant is that using a language model is like doing air guitar.

AUDIO: DK theme

Maybe we thought the infantilization of adult life had peaked with video game championships, pardon me, e-sports; well, we were wrong. What for my generation ended with the discovery of non-individual sex is now a profession.

And not to miss anything there are, and how could they miss, air guitar championships. Maybe someone thought that wiggling in time to music while pretending to play an imaginary guitar is one of those two or three things that should be confined to when one is taking a shower and no one is around; but no.

What does air guitar have to do with it? Everything is deeply interconnected, Ted Nelson always says.

The other day I read a bunch of articles about programming with various intelligent assistants. I read about the one who calls his skeptical friends imbeciles; I read Kenton Varda, who in CloudFlare developed an OAuth authentication library using Claude from Anthropic.

I read myself Neil Madden, an OAuth expert who grasps at straws not to say he would burn it all down (because being possibilistic about AI is the right career move anyway) but then concludes:

For trivial things, which I don't care how they're done, I'm happy to let a language model do whatever it wants. But for important things, like my fucking authentication system, I'd rather do it myself and be sure I've really thought about it.

I would like to see what exactly are the parts of code where one can say "I don't care how they are done." I swear, I'm curious. But that's not the point.

The point is that when we talk about programming with AI assistants, we are talking about programming as managers and stakeholders understand it.

Let me see if I can explain myself.

I don't know if you've happened to see a pianist play recently. Those fingers flying over the keys, that almost alien naturalness.

Yet, if I were to tell you that I want to study piano to improve the agility of my fingers you would tell me that I am wasting my time.

Or have you happened to see a programmer writing code. There too, fingers flying across the keyboard, lines of code flowing almost like a Hollywood movie.

Let's ignore for a moment that the most frequently used key on the keyboard is Backspace, and that physically writing code takes up perhaps one-fifth of a programmer's time.

But if I were to tell you that I measure a programmer's ability by the speed with which he or she writes code, there are two possibilities:

1. you are not in the industry, however, you feel that something is wrong because you are smart enough to breathe
2. you are in the industry, and you rightly call me a moron.

Unfortunately, there is little to laugh about. The number of lines of code produced was really a metric in my early years of work. It didn't last long, because it's like measuring how good a pianist is by the number of keys pressed in a minute. It is nonsense. Notice that in the case of the pianist I didn't say "speed of execution," which at least might give us a freak show, I really said "number of keys pressed," whatever they are.

Here, measuring a programmer by the amount of code written, or the speed with which it is produced, is another idiocy.

In the early days of computing, machines had such enormous costs, that software was not even considered a product in itself; and no one questioned whether a certain program took 5 or 10 man-years.

But the moment digital technology began to be mainstream, and hardware to cost less and less, programming became the cost to contain.

And therein lies the problem. Because programming is an extremely complex intellectual activity, most of which is... reasoning. The point is that reasoning cannot be seen, much less measured.

When a programmer looks out the window for half an hour, or fills three blackboards with notes, he is working. The code comes only downstream from a whole series of reasoning and simulations, assumptions, confirmations and denials, that the programmer has already done in his mind. This does not mean at all that the code is correct; it means that that particular code is preferable to dozens of alternatives that the programmer has already ruled out because he already knows they are wrong.

Because of this, decades of so-called innovation focused on the speed with which code is written have produced two definite results:

• more poorly written code, and
• more keystrokes and less thinking.

Then we can argue about changes in the quality of software, if we want to cry.

But since, precisely, the reasoning is not seen, and the code is, managers, the software industry, and of course Hollywood, to whom we owe the public's idea of what programming is and how a programmer behaves, have focused on what is seen instead: the code.

From this perspective, language models are only the latest step in a long journey that has reduced

• the computer science profession to a digital proletariat
• and the methodologies (or if you will, the discipline) of development to trial and error, plus tickets in Jira and diagrams in miro.

My first job was to develop a multimedia hypertext platform for the knowledge worker. After thirty-five years of the so-called "information society," the knowledge worker is now doing SEO, social media manager, piecework crusher, or even seeing his or her work resold as an AI product, like the Nigerian data labelers or the seven hundred Indian engineers of the startup "builder.ai," which raised hundreds of millions in investment before anyone realized it was a scam.

(Interestingly, builder.ai filed for bankruptcy but no one was sued for scamming. Investors don't care to let people know how mouthy they are).

But what I've seen in IT, LLMs are extending to the entire intellectual sector.

While the whole education sector is already creaking in the face of on the one hand students who instead of learning write prompts, and on the other hand faculty who have chatGPT correct tests, the professions sector is no better off.

Coaches, psychologists, teachers, doctors, lawyers... I can also understand Altman and company going around blathering that "in a few years" (it was in a few years even when he started, in fact, three years ago it was "in five years"). They have to be understood, they have to sell, and if all you have in stock is cow dung, cow dung has unexpected healing properties.

What I don't understand, and what saddens me, is to see so many professionals, even capable ones, buying into this egregious hoax and even supporting it.

Without understanding that the sole purpose of LLMs is to lower the market value of human skills.

Deluded professionals believe that the genie in the bottle of language patterns will turn them into managers, and they do not understand that they are undermining their own credibility.

And the deluded professionals are, of course, joined by torments of users, who are also deluded into thinking they are becoming professionals by that other hoax, of the "democratization" of the professions.

But there is no democratization. Competence is not a problem of accessibility or democracy. It is an issue of hard work, investment of time and money. Looking for shortcuts, it just means cheating and not expecting consequences.

Managers, shareholders and deluded professionals see in language models the possibility of having the work (or at least the visible and saleable aspects of the work), but without the workers.

Deluded users see in it the possibility of having the spectacle of competence, without the competence.

One and all, they play the role that Lenin (this time I'm sure it's Lenin) would have called of "useful idiots." All wiggling like obsessives with their air guitars, thinking they are David Gilmour and expecting writing.

With the added problem that these useful idiots play not for the revolution in the class struggle, but for their own oppressors.