Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Ceiling is being raised. cursor's copilot helped us write "superhuman code" (ai88.substack.com)
12 points by uliyahoo on May 30, 2024 | hide | past | favorite | 19 comments


>> Ironically, the whole point of this scary complex “superior mega brained” type code is… to let the “anti-typists” have their cake and eat it too... We realized that we can provide all of the safety + convenience benefits of Zod to those who choose plain JSON schemas - with zero overhead for users of the framework.

How did we get to a place where this is what modern development looks like? Does this seem completely off the fucking rails to any one else? I feel like we have all gone the full Patty Herest and embraced the Stockholm syndrome at this point... "Well Im not gonna be able to fix it so I might as well power up the problems".

Our industry needs an enema...


Calling something this simple “superhuman” is more an indictment of the author’s skill level than anything else.


It is bad code actuallly.


Said the human worker just before he was fired, never to be heard from again. Meanwhile, the code kept working correctly for the next ten years.


Click into the blog, the code is surprisingly good and actually pretty complicated.


I don't work with typescript much but that code is pretty straightforward, it's just mapping objects to types. A few nested conditions and recursive calls aren't complicated at all. That a decent amount of devs seem to think it is makes me confident in my job security, at least.


Why on earth would you assume I hadn’t read the code? It’s fine, not particularly complicated, and certainly not superhuman.


I guess I shouldn't have assumed that


Where are you gonna get some superhumans to maintain that code down the line, though?


For that we'll need AGI...


That doesn’t seem super human, or even very complicated. I think I’ve written more complex code to summarise output from instrumentation data within the last couple of weeks, and that was when I was on a plane without access to the internet to look stuff up.


The real danger of AI.

When people become habituated to getting a mostly correct result most of the time they will start excepting AI generated answers, code etc as "correct" despite not being able to fully understand what it is doing. Understanding what circumstances the AI might do something wrong or life threatening will eventually be classified as an acceptable risk. Given the amount of 'sounds good but is actually complete gibberish' websites I see that are obviously AI generated this may have already happened.


So the meta-cognition problem has been solved?


Poor guy doesn't have macros.


I remember that X post, pretty funny


It got wrecked...


Sorry but understanding code snippets is not what makes maintaining old code difficult. Its understanding why it is the way it is within the context of the rest of the code base. If something is so completely decoupled an AI can futz with the code with out any problems within the rest of the code base a human can also just toss the code and write new code.

AI doesn't help us maintain large code bases still.

Programming as theory building!


I see "Q/acc" in the authors twitter name, which I guess is a sign of allegiance to some modern Silicon Valley movement or ideology like how "e/acc" is Effective Accelerationism. Anyone know which one this is?


I think it’s a cryptocurrency, but it’s quite hard to tell.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: