By an order of magnitude, the most valuable use case of LLMs has been coding.
It’s another shift in our history similar to the leap from painting to photography.
Discussion
I guess I'm the only idiot who respects and admires the old oil painters who spent their entire lives mastering the craft...there's just something about the old days of writing code organically by reading good documentation, not to mention the true masters who could write beautiful and maintainable code ...ahh nevermind (turns to dust and blows away in wind)
Same applies to photography. You need millions of images taken, hundreds of hours spend on researching, planning and developing, and thousands of miles traveled, most of the times in vain. Just to do it properly.
One is creation, the other is capture, kinda similar to AI training off what already exists, actually. I do think photos can be creative and artistic but not sure if it's the same. Degradation of art is a thing, IMO
It’s a different art whatsoever. If the art of painting is deciding what to include, the art of photography is deciding what to exclude. It’s dumb to compare the two.
One does not exclude the other. Photography didn’t replace painting, it expanded the art of telling stories through images to millions more people and new mediums.
Coding with LLMs will do the same for a different domain.
The way I see it is like the digital camera or cellphone camera allowing the masses to snap a photo of anything and creating a giant archive of disposable slop. This will be the same with LLMs and code for most of the users. Bringing a technology to the masses doesn't mean everyone will be generating good quality results.
The cream will rise to the top of the new "art" and we will be left with amateur "devs" posting their LLM selfie to LLMstagram.
When the hype dies down and everyone starts being critical of software and code again it will be obvious.
Of course there will be new slop, that's actually the proof that it has created a paradigm shift in how technology is used and delivered.
We saw the same thing with Youtube, there was an ocean of worthless, nonsense content that nobody wanted to watch. But the existence of that in a way that it could be interacted with by millions of people proved that it was truly a fundamental shift in media and publication, and now those types of platforms have content that is exclusive to their medium and types of productions that would literally never have existed otherwise.
I expect to see the same type of change when it comes to people building software with LLMs, enormous amounts of slop, but an entirely new experience and new ideas that only became possible because of the millions of new people able to execute on an idea and bring it into reality.
I hope you're right, Guy, cause I like ya and think you're a sweetheart with a fat sat account!
With that, I'm going to put aside the other software dev methodologies that came about over the years and how they only made more and more slop. See: Agile. The good stuff always comes to the forefront no matter the Next Big Thing...
Also, I don't want to conflate vibe coding with LLM auto completion and boilerplate either, which is actually useful. In the above I'm only speaking on the art of vibe coding.
The art of painting is deciding what to include in it. The art of photography is deciding what to exclude.
Yea dude..I've learned a ton of html and python using chat gpt..it fucks up from time to time tho..but still amazing
agree but most people dont use it for that.
general queries that used to require digging through pages of search results and ads seems like a bigger impact in terms of scale
That has been the ah ha for me. Quick question, send prompt, get a result, and move on.
It will make average coders good, good coders great, and bad coders worse. It’s kind of brilliant.
I am using it to replace myself, it is doing a pretty good job so far. A few years and I will be obsolete ☺️