Hacker News Reader

Comment by rowanseymour | Hacker News Reader

> Programming is getting democratized

I have no idea what that is supposed to mean but I keep hearing the same about art, music and other creative fields and it sure sounds like contempt for creative people.

I personally don't lose any sleep over LLMs being powerful wizards for getting started on new projects.. that's what LLMs are good at.. pulling together bits of things they've seen on the internet. There's a chasm between that and maintaining, iterating on a complex project. Things that require actual intelligence.

View on HN

It feels to me like it's getting democratized in the same sense as to what happened to professional photography in the early 2000s with the introduction of digital cameras and high quality color inkjet printers. The barrier to entry becomes so much lower.

Instead of dealing with the costs associated with using, developing and printing from film, as well as the skills associated with knowing what a photo would look like before it was developed, digital cameras allowed new photographers to enter the industry relatively cheaply and shoot off a few thousand photos at a wedding at a relatively negligible cost. Those photographers rapidly developed their skills, and left studios with massive million dollar Kodak digital chemical printers in the dust. I know because I was working at one.

If you remember, this was in the time where the studio owned your negatives ostensibly forever, and you had to pay for reprints or enlargements. What were amateur photographers could enter this high-margin market, produce images of an acceptable quality, charge far less and provide far more.

I'm not able to say whether this will happen to software development, but the democratization of professional photography absolutely shook the somewhat complacent industry to its core.

In that case it had nothing to do with contempt for creative people, it was the opposite, anyone who wanted to be creative now could be.

Digital cameras didn't change the need to go out and actually shoot the photos, however. They didn't change the fundamentals of lighting and color and what a good photo looks like. It was a more convenient and cheaper process compared to film, so more people could participate in photography, but it maintained a lot of the creative process.

It means that you no longer need to go through an intermediary to get the result you want. You have the end product in your mind, you just don't have the ability to translate it into reality.

I can give you the real example of recently needing to translates ancient 90's era manufacturing files into modern ones, while also generating companion automation files from it (which needs to by done manually but with tooling to facilitate the process).

I found a company that sells software capable of doing this. A license is $1000/yr/usr.

The next day I was able to get claude 3.7 to make a program that does all that, translate the files, and then have a GUI that renders them so an engineer can go through and manually demarcate points, which then are used to calculate and output the automation file. This took about 45 minutes and is what the department is using now. We have thousands of these files that will need to get modernized as they get used.

I see this everywhere now, and I have been building bespoke programs left an right. Whether it be an audio spectrum analyzer that allows me to finely tune my stereo equalizer visually based on feedback for any given track, or an app on my phone that instantly calculates futures prices and margin requirements for given position sizes and predicted market movements.

People think LLMs will be a paradigm shift, I can tell you the shift has already happened, it's just a matter of awareness now.

> an app on my phone that instantly calculates futures prices and margin requirements for given position sizes and predicted market movements.

That sounds like something for which one should be spending the money on professionally developed and well-tested software. What's the expression? Penny wise, pound foolish.

Not really, because the equations are straight forward and not particularly difficult. Most people could model it in excel no problem. Is it a guarantee that it is robust? No. But I'm also only using it myself instead of ball parking like I previously would.

My sister who has never coded in her life is building mini apps to do things at work.

It’s taken her a month to get up and running.

Everyone needs to realize this is not a matter of AI stealing dev jobs but the supply of low skilled developers exploding and eating into the market.

In this case, "democratized" means "enshittified".

https://pluralistic.net/2023/01/21/potemkin-ai/#hey-guys

I don't doubt that LLMs can make programmers more productive. It's happening today, and I expect it will continue to improve, but it requires knowing what code they should generate, what the actual goals are, and how it should be tested. It can generate standard solutions to standard problems with standard bugs. That's fine they're a tool.

What the inexperienced expect them to do is read their mind, and implement what they want without testing (other than did it crash the first time I used it). Unfortunately, knowing the questions to ask is at least half of the problem, which by definition the inexperienced don't know how to do. You can already see that with vibecoding prompts to "write clear comments", "don't write bugs", and "use best practices".

So why does it lead to the enshitification of the programming experience? Because regular folks will be led to believe (Startrek movie Wargames hacker style) that this is how things are done. The will accept and expect rehashed garbage UI and implementations without security or corner case checking, because that's what they always get when they press a button and wait a minute. Now, why can't YOU stupid programmer, get the same results faster? I told you I wanted a cool game that would make me lots of money fast with no bugs!

I do have hope that some people will learn to be more clear in their descriptions of things, but guess what, english isn't really the language for that.

I'm not talking about programmers using LLMs.

I'm talking about people talking in english to an AI on one screen, and compiled functioning programs appearing on the other. An "app playground" where you just tell your phone what you need an app to do, and a new bespoke app is now in your app draw.

Forget about UIs too. They won't be that important. You don't need a tree of options and menus, tool bars and buttons. You would just tell the program what you want it to do..."Don't put my signature on this email"..."wrap the text around this image properly"(cough msword cough)..."split these two parts and move the red one to the top layer"...or even "Create a button that does this and place it over there".

I'll accept that you could get an IfThisThenThat implementation that might let you get some home automation done. Of course, if you can't build an ITTT to start with, you'll have a horrible time debugging it, or even remembering to turn it off. The create a magic button thing though is exactly the terrible UI I'm talking about.

I think part of what you want is voice applications, because deleting your signature by hand is probably easier than trying to build a program that does it. Maybe the app could just search help and tell you what feature already does what you're asking for. Certainly, context sensitive voice recognition has gotten a LOT better with the latest LLMs. Not sure I'm looking forward to the guy on the train narrating to his laptop for an excel page, though.

But when art is getting democratised, it is stealing, like we learnt from Ghiblification incident lask week.

AI isn't democratizing art. Art has always been democratized. If you want to draw something in the Ghibli style, go ahead, no one is stopping you. I saw plenty of great Ghibli style non-AI art on my timeline in response to that, some by people who just spent a few days teaching themselves. The means have never been more accessible.

But using AI to generate something in that style doesn't make you an artist. It isn't art, it's just a product.

> But using AI to generate something in that style doesn't make you an artist. It isn't art, it's just a product.

This is probably how people felt at the advent of calculators

Equating the process of creating art with adding numbers together is a very HN answer. Unsurprising that so few people here see value in human expression, when their only concern is efficiently generating a minimum viable product as quickly as possible.

But... using a calculator doesn't make you a mathematician either. And one could argue that society has born real negative consequences from the inability of most people to do even basic math because of the ubiquity of calculators. There is a big difference between using a tool and having the tool do everything for you.

Do you really believe that society will benefit when most people don't know how to express themselves creatively in any way other than asking the magic box to make a pretty thing for them?

Calculators, abacuses etc are tools that let you do math easier than in your head. People have done original math on all sorts of things.

Generative AI forces us to reconsider what original means because it's producing a "remix" of what it has seen before with no credit going to those who created those original works.

Pretty much my exact sentiments. The Internet for a long time has contained more than enough information and connections to other people for anyone to learn nearly any skill they want. Anyone who wants to put in the work has the opportunity.

Celebrating the 'democratization' of these skills is just showing adversity to basic learning and thinking. I'm not gonna celebrate a billion dollar corp trying to replace fundamentals of being human.

>Celebrating the 'democratization' of these skills is just showing adversity to basic learning and thinking.

The reality is that you cannot become an expert in everything. I have songs I'd love to compose in my head, but it would be totally impractical for me to go through the hundreds/thousands of hours of training that would be needed to realize these songs in reality. Nor am I particularly motivated to pay someone else to sit there for hours trying to compose what I am telling them.

This is true for hundreds of activities. Things I want to do, but cannot devote the time to learn the intermediate steps to get there.

>I have songs I'd love to compose in my head, but it would be totally impractical for me to go through the hundreds/thousands of hours of training that would be needed to realize these songs in reality. Nor am I particularly motivated to pay someone else to sit there for hours trying to compose what I am telling them.

So the alternative is that you'll pay a tech company instead -- to use their model trained on unlicensed and uncredited human works to generate a mishmash of plagiarized songs, the end result of which nobody will ever want to listen to?

> you cannot become an expert in everything

You don't have to though. Anyone who's spent a decent amount of time in a creative hobby will tell you they sucked when they started but they enjoyed the process of learning and exploring. I think you're depriving yourself of the mental benefits of learning a new skill and being creative. It flexes your mind in new ways.

If you just want something to exist, sure, but when you can press buttons and have a magic box spit out whatever you want with no effort, how much are you actually going to value it?

If you only knew how many hobbies I have already lol

It's not stealing when the original still exists. And art style is not copyrightable.

Our current laws are made to handle AI

> There's a chasm between that and maintaining, iterating on a complex project.

The can for sure maintain and iterate on smaller projects -- which millions of people are happily feeding into them as training data.

Going from a project with 1,000 to 1,000,000 lines of code is a tiny leap compared to going from 0 to 1000. Once your argument is based on scale then you pretty much lost to the robots.

I'm not saying they are going to invent some free energy source (other than using humans as batteries) anytime soon but when the argument is "you're going to be blindsided" and the response shows, and I'm not trying to be insulting or anything like that, willful ignorance of the facts on the ground I'm going to say you probably will be blindsided when they take your app writing job or whatever.

I'm not attacking you at all just saying that there's a bunch of people who chose to keep their heads in the sand and hope this all just goes away.

I'm sure there are lots of engineers who don't want to change but I don't feel I'm one of those. I use copilot every day. I'm leading integrating our product with different LLMs and playing with those regularly.

And yet I feel more secure in my job today than I did a year ago because I'm constantly hitting the limits of what a language model can do. I've realized that the decisions I make every day as a senior engineer aren't mostly about what lines of code usually come after each other.

> Going from a project with 1,000 to 1,000,000 lines of code is a tiny leap compared to going from 0 to 1000.

Are you sure the leap is tiny? It's a much easier problem to get only 1,000 lines of code to be correct, because the lines only have to be consistent with each other.

Right, and projects that are millions lines of code usually aren't even one codebase or one app or service.

Reply on news.ycombinator.com