Photography is a great example here, because it seems pretty possible that generative models do to writing what smartphones did to photography.
People who want to take pictures no longer need to be photographers.
I think it’s interesting to look at, but I think the parallels break down pretty quickly. The decline of legacy photographic equipment and the reduction of skill needed to take photos is all about a rather technical skillset becoming less and less necessary. The smartphone also made many other legacy things irrelevant for most people.
Language and writing are closer to the core of humanity in a rather existential way. It’s less clear that reducing the skillset required to participate in writing is a net benefit, even if it enables the generation of more text by people who might not have done so before.
Smartphones make a photographer’s job easier and a non-photographer’s goals possible. I worry that the same can’t be said for language models and the end result is likely to be closer to a new form of illiteracy than it is to the democratization of a process that formerly required greater degrees of technical skill.
When painting broad strokes, I see the similarity. Drilling down even a little widens the conceptual gap quite a bit.
It's easy to imagine the same thing happening with text. In fact, the speed that lawmakers and judges adopted the idea of generated text is a very big concern to me.
I sometimes hear people say „there is so much interesting stuff to watch on youtube/pinterest/whatnot“ and i always wonder how the vastly the meaning of „interesting“ can vary.
Smartphone cameras are great and easy to use for a very narrow set of conditions. Outside those conditions they perform terribly. The kicker is that you don’t really know unless you already know… the “trained eye” can immediately notice.
The same is true for GPT text/code.
Even then, yes, it will still result shifts like we saw years ago with newspapers letting go of their photo staff and telling journalists to use their phones.
I know what they did to cameras, but photography?
You still need the person who takes the picture, and the process of taking a picture hasn't really changed that much.
I don't know why LLMs would do that with "writing" though.
The same might potentially happen with writing, where if anyone can generate a decent original story, poem, advertising copy, product UI writing, etc. in less than five minutes, the lowest and middle tiers of paid writing work will be eliminated.
It’s possible that because these writing-specific jobs were already low value / largely eliminated by the Internet at large, that won’t happen. But it could happen to almost any role that basically involves synthesis and summary of words and data, since LLMs are so general purpose.
But digital cameras for the everyday person existed long before the smartphone. And before that, point and shoot film cameras and Polaroid.
The marginal cost of taking a photo is now $0. It's a hobby that anyone with a smartphone can now take up.
It’s was already a near universal hobby.
20 years ago nearly everyone took pictures at some point. Every house had a camera or two, and disposal cameras were cheap and very widely available.
It was widely regarded as a universal hobby back then. My mom was a wedding photographer, and people would regulars make the same arguments back then.
If you look at historic BLS data the number of photographers has actually grown faster than the population has since 2020, and the wages in 2021 were about the same adjusted for inflation.
Clearly smartphones haven’t had a huge impact on job numbers or salaries. My hypothesis is because photographic skill was already as devalued as it was likely to get, and that actual technical skill is a very small part of the overall job.