Lots of things are find when they are first introduced because we carry our li ed experiences into it. But those who grow up with the new thing shaping their lived experience don’t bring that perspective to it.
I am an embedded developer so the LLM/AI is a omni-present talking point and one of my friends was saying that he loved LLMs because they could generate big chunks of code and he can go through it after and fix the mistakes.
He has the skills to fix the issues because he has a decade of non-LLM experience.
But someone that doesn’t have that experience will have a hard time finding the correct fix when the vibe code isn’t working as it should. They will rely on LLM telling them that they were right, so here is a new fix that doesn’t fix the issue.
I have such a hangup on this. Currently, a “tech journalist” in one of the big newspapers in my country is doing a series of articles about how he’s vibe coded an app that, apparently, has been green-lighted by the IT department and is very useful for his fellow journalists.
He admits to not being able to read or write a single line of code, and describes what he does as “leading a team” where he makes decisions about what kind of features to implement, when things are too slow and need speed improvements, etc. Apparently, this web-app is now 66 000 lines of code, and used in production (unclear what it’s actually used for). The LLM agents take care of everything from writing the code to setting up PR’s, reviewing, testing, and deploying.
I can’t help but see so painfully clearly that he’s created 66 000 lines of liability, that he has exactly zero concept of potential bugs in, and which no human in the world is likely to fix quickly if production goes down. He has no idea whether database rollbacks are safe or even possible if something is corrupted… there’s just so many foot canons waiting to go off. And this is just 66k lines. That’s not even a small web-app, it’s tiny (this guy can’t see the difference between generated files and written files, so I’m assuming 66k includes everything), and my personal experience is that LLM agents just get worse as complexity increases.
The biggest problem is that it’s painfully clear that this guy is oblivious to all the above. He’s happily chugging along as long as this looks like it’s working. I can only assume that other people with his level of experience (that is, none) see it the same way.
I agree to a certain extent, but also would point out that the longer something exists, the more likely it is to be exploited by capital and then just be objectively worse. A lot of the good things are still there, just pushed aside and hidden by people with ulterior motives. Plenty of queer youth still find a lot of community and comfort online when it’s absent in their lives. There is definitely something to be said about how things shape people vs how people shape things though.
Lots of things are find when they are first introduced because we carry our li ed experiences into it. But those who grow up with the new thing shaping their lived experience don’t bring that perspective to it.
Thanks for putting it in such concise words.
I am an embedded developer so the LLM/AI is a omni-present talking point and one of my friends was saying that he loved LLMs because they could generate big chunks of code and he can go through it after and fix the mistakes.
He has the skills to fix the issues because he has a decade of non-LLM experience.
But someone that doesn’t have that experience will have a hard time finding the correct fix when the vibe code isn’t working as it should. They will rely on LLM telling them that they were right, so here is a new fix that doesn’t fix the issue.
I have such a hangup on this. Currently, a “tech journalist” in one of the big newspapers in my country is doing a series of articles about how he’s vibe coded an app that, apparently, has been green-lighted by the IT department and is very useful for his fellow journalists.
He admits to not being able to read or write a single line of code, and describes what he does as “leading a team” where he makes decisions about what kind of features to implement, when things are too slow and need speed improvements, etc. Apparently, this web-app is now 66 000 lines of code, and used in production (unclear what it’s actually used for). The LLM agents take care of everything from writing the code to setting up PR’s, reviewing, testing, and deploying.
I can’t help but see so painfully clearly that he’s created 66 000 lines of liability, that he has exactly zero concept of potential bugs in, and which no human in the world is likely to fix quickly if production goes down. He has no idea whether database rollbacks are safe or even possible if something is corrupted… there’s just so many foot canons waiting to go off. And this is just 66k lines. That’s not even a small web-app, it’s tiny (this guy can’t see the difference between generated files and written files, so I’m assuming 66k includes everything), and my personal experience is that LLM agents just get worse as complexity increases.
The biggest problem is that it’s painfully clear that this guy is oblivious to all the above. He’s happily chugging along as long as this looks like it’s working. I can only assume that other people with his level of experience (that is, none) see it the same way.
I agree to a certain extent, but also would point out that the longer something exists, the more likely it is to be exploited by capital and then just be objectively worse. A lot of the good things are still there, just pushed aside and hidden by people with ulterior motives. Plenty of queer youth still find a lot of community and comfort online when it’s absent in their lives. There is definitely something to be said about how things shape people vs how people shape things though.