Login
You're viewing the functional.cafe public feed.
  • Jun 20, 2026, 12:55 AM

    I can attest with the utmost confidence that every developer or manager or whoever that claims "all code written by AI is reviewed by a human" is a boldfaced lie.

    AT BEST they have quickly skimmed it, but more likely they've not looked at it at all.

    Actually reading code THAT YOU DID NOT WRITE YOURSELF takes quite a bit of time, and the one thing that people who use AI tooling care most about is just how fast it makes things.

    They will not spend the time needed to actually read it.

    💬 8🔄 69⭐ 3

Replies

  • Jun 20, 2026, 1:00 AM

    Any developer worth their salt will know that bugs in code arent typically obvious, and the severity of a programming mistake has literally no association with how much code relates to it.

    Single-character bugs can be disasters.

    You cannot reliably catch a single character being incorrect when glancing through code, especially code you didn't write, ESPECIALLY code you're unfamiliar with.

    💬 1🔄 11⭐ 0
  • Jun 20, 2026, 1:02 AM

    So when somebody tries to defend their use of these tools because a human is still reviewing it remember that they're lying - to you and to themselves.

    They know there's problems with these tools, and are trying to convince themselves that somehow it's fine and you should stop calling out for their poor choices.

    💬 4🔄 12⭐ 44
  • 💬 5🔄 16⭐ 40
  • Jul 3, 2026, 9:58 PM

    @ecn @davidgerard nobody has demonstrated a method for combating vigilance decay yet, and we really need to stop all usage of these tools at scale immediately until someone does

    💬 2🔄 3⭐ 33
  • 💬 0🔄 0⭐ 7
  • Nov 12, 2026, 3:24 AM

    @glyph @ecn @davidgerard I am constantly reminded of how one of the most important things when studying is *writing things down*.  Physically writing it down (not sure if a keyboard is the same but) supposedly allows you to remember more of what you've written down.

    https://pubmed.ncbi.nlm.nih.gov/21506451/ just one random ass link, I am sure a psychology cat will *read my ass to filth*, but basically you're engaging with the material more when you're doing things like "writing down and taking notes" rather than just reading or listening and that seems linked to memory and/or understanding.

    So if you're just *skimming* text output dumped out by a machine, where in the world is the understanding supposed to come in??  I don't think just skimming a textbook WITHOUT solving problems and taking notes has EVER WORKED FOR ANYONE.  Hell, even when I've taught as a TA in freshman chemistry, I always have to go back and re-work the problems to remember what I'm doing and work around the subtle gotchas.

    The idea that *somehow* code is different in this regard is fucking preposterous and an extraordinary claim that requires, frankly, extraordinary evidence.  Just add that to the pile of those, I suppose...

    💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 8
  • 💬 1🔄 0⭐ 5
  • 💬 0🔄 0⭐ 5
  • 💬 1🔄 0⭐ 0
  • Jul 4, 2026, 9:31 AM

    @TheOneDoc @Rhube @glyph @ecn @davidgerard

    When you use "Delusion" which is as you say is a Diagnosis, like a swear word or to lessen people, which might not have this diagnosis, that that is clearly Ableism.

    You basically say at the same time "People with Delusion/Insanity are so dumb, they would....", when you are saying "Unbelievable how dumb these AI People are, they must be Delusional and Insane"

    Please take such accusations seriously and not as personal Insult. Thank you :)

    💬 1🔄 0⭐ 2
  • Jul 4, 2026, 11:00 AM

    @kb01 @Rhube @glyph @ecn @davidgerard how about fiuck you?

    I diagnosed those people with a specific kind of delusion and as long as you don't show me a well reasoned diagnosis that disagrees with mine I stand by what I said.

    Sag mal hältst du mich für dämlich?

    You basically say at the same time "People with Delusion/Insanity are so dumb, they would....",

    I said what I said not what you want to read into it.

    Those people are not only a danger for themselves they are one for all of humanity.

    Now go and white knight for people who need it and not execs that gladly let you starve to death to make one more buck.

    We don't need an other protector of the protected.

    plonk

    💬 0🔄 0⭐ 1
  • Jul 4, 2026, 2:55 PM

    @TheOneDoc @Rhube @glyph @ecn @davidgerard
    When did it became normal to diagnose groups of people in the Internet?
    What gives you the right to do so?
    What is your qualification to do medical Diagnosis?

    Also why should I also diagnose AI Users, to argue if something might be Ableism?Ableism
    I won't argue for a medical Diagnosis, not for AI Users, not for you, not for anybody, because it just feels sooo wrong!

    💬 0🔄 0⭐ 0
  • Jul 4, 2026, 10:42 AM

    @glyph
    Anyone with a touch of self awareness knew this, which says much about those who didn't expect this effect, or just choose to ignore the issue.

    So much of what's really bad about #LLM tech, as well as it's limitations is obvious to many, but here we are.
    @ecn @davidgerard

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Nov 12, 2026, 7:52 AM

    @glyph @ecn @davidgerard

    Hugging Face published a paper making the same point a couple of weeks ago.

    It shouldn’t exactly be news, the Paradox of Automation has been known for a very long time. It was one of the motivations for the big shift in Japanese engineering process in the ‘80s. And it predicts exactly the outcome that these companies are now discovering.

    But having Amazon and Hugging Face publicly say it is nice.

    💬 1🔄 1⭐ 0
  • 💬 0🔄 0⭐ 0
  • Jul 3, 2026, 10:54 PM

    @ecn I contributed to the Drupal open source project for 22 years. Every Drupal core commit is reviewed by smart, hard working programmers who truly give their best. And yet, Drupal core has bugs and security holes. Something to think about.

    💬 0🔄 0⭐ 0
  • Jul 4, 2026, 12:10 AM

    @ecn I’d go a step further and say that, AI generated code is *especially* hard to review.

    Let me explain.

    After several years of reviewing human written code, one develops a sort of 6th sense that allows them to identify potentially problematic areas by skimming through the code. Once a suspicious area is encountered, the experienced reviewer will then examine it throughly and decide wether it was a false alarm or an actual problem.

    AI generated code is hard to review because “it always looks good” while skimming. So one has to carefully examine every component to find problems, and problems are usually very subtle. It’s so tiring, honestly.

    At least that has been my experience so far.

    💬 0🔄 0⭐ 3
  • Jul 4, 2026, 6:10 AM

    @ecn Isn't code review about architecture and maintainability, and not about finding bugs? There is the entire testing process to find bugs. Sure, you could find a bug by seeing an obvious one, I just don't think thats the main goal of a code review nor that many people can "see" performance bottlenecks or race conditions by just reading code.

    💬 0🔄 0⭐ 1
  • 💬 0🔄 1⭐ 4
  • 💬 0🔄 0⭐ 0
  • Nov 12, 2026, 2:27 AM

    @ecn Half the time I don't actually understand the comments that have been written^H^H^H^H^H^Hcreated, well before I even look at the code.

    If you can't meaningfully and succinctly describe what the code is *for*, for a reviewer, how can they review whether it even achieves its intended goal?

    💬 0🔄 0⭐ 0
  • Nov 12, 2026, 5:24 AM

    @ecn 🖐️ I’ve used AI to generate code for personal projects (never for shipping code at work, of course) and I read every line of output to made sure I understood exactly what it was doing and often rewrote it to fit better in my application. Just wanted to say hi. Cheers!

    💬 1🔄 0⭐ 0
  • Nov 12, 2026, 5:29 AM

    @ecn I find it useful for simple stuff that’s mostly boilerplate, motor drivers and the like. Much more complex than that and you spend more time arguing with it than you save

    💬 0🔄 0⭐ 0
  • Nov 12, 2026, 8:40 AM

    @ecn

    Indeed. I don't know anyone who wants to spend time reviewing AI generated code.

    The human-in-the-loop narrative was meant to be transitive towards a fully agentic / autonomous process from the start.

    💬 0🔄 0⭐ 0
  • Nov 12, 2026, 9:09 AM

    @ecn Even if they start with the best of intentions, they they will definitely properly review all the AI written code, that plan will not survive the daily reality of time pressures and natural human laziness and complacency. It's the easiest part of the process to skip. So it will be skipped.

    💬 0🔄 0⭐ 0
  • Nov 12, 2026, 10:23 AM

    @ecn
    Some years ago in a city I visit often, the public transit operator was taken over by growth-hacking tech fucks. One of their "innovations" was a quasi-mandatory app-payment system where the app shows an animated pattern that bus drivers are supposed to check when passengers enter (there is a QR code but only inspectors check that).

    So of course someone just did a webapp which generates the same animation with today's date on top.

    One of the top "solutions": punish bus drivers for not examining people's phones in detail.

    The AI bro "human in the loop" shit always makes me think of this system.

    💬 0🔄 0⭐ 0