Login
You're viewing the discuss.systems public feed.
  • Hazel Weaklyhazelweakly@hachyderm.io
    Oct 5, 2026, 4:18 PM

    I don’t know how to explain to people that just because certain things appear to approximately work doesn’t mean they actually work. This was already horrifically difficult before AI and now it’s damn near impossible to get the point across when literally any trivial example passes muster

    The asymmetry between me having to provide a compelling defence that they can actually understand (without the required background knowledge) vs them being able to endlessly generate garbage in seconds is just… exhausting

    💬 15🔄 123⭐ 166

Replies

  • 💬 0🔄 1⭐ 9
  • Chris Swancpswan@hachyderm.io
    Oct 5, 2026, 4:55 PM

    @hazelweakly It's all very tedious, but the DevOps movement did prepare some ground for us.

    T: "I built a thing (with my magic coding assistant, I can't explain it, which is why it's magic)"
    Y: "neat - 'you build it, you run it'"
    T (a while later): "the magic coding assistant said stuff that I don't understand"
    Y: "that sounds like a you problem rather than a me problem"

    Or for the proper grown up version we can play by SRE rules...

    T: "I built a thing (with my magic coding assistant, I can't explain it, which is why it's magic)"
    Y: "OK, let's agree some SLOs and error budgets, and the underlying metrics that we'll track to feed those"
    T: "why are you making me do work?"
    Y: "so that we both know when it's working?"

    💬 2🔄 7⭐ 14
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 1⭐ 2
  • (roll m3tti)m3tti@functional.cafe
    Oct 5, 2026, 6:36 PM

    @hazelweakly best ai story from a big corpo i worked for. We have to do something in cloudflare (oh no not this guys) but anyway. How do we solve it. We can do it with some js code and deploy it there. Yeah sure good idea. Chatgpt said use this code for the cloudflare worker. Looks into code finds document.getElementById and window. Now How do i explain to them that this code couldnt work without deploying in the first place.

    I love ai [/sarcasm]

    💬 0🔄 0⭐ 1
  • Lady of Mystery and Sciencedemi@xeno.glyphpress.com
    Oct 5, 2026, 7:50 PM

    @hazelweakly
    For literally decades I’ve been trying to get people to understand that: just because something looks like it’s working correctly does not mean that is in fact working correctly.
    And the example I always use is: Is a stopped clock actually right twice each day? Or does it just _look_ right twice each day?

    💬 0🔄 0⭐ 3
  • 💬 0🔄 1⭐ 0
  • Kevinkhwevin@neuromatch.social
    Oct 5, 2026, 8:46 PM

    @hazelweakly well said... And a bit sad.

    Tech Bros talk about AI as a catalyst, an accelerator, something that multiplies potential... Seems like our shit potential is sky rocketting.

    💬 0🔄 0⭐ 0
  • Yvan ー イボん 🗺️ :ferris: :go:YvanDaSilva@hachyderm.io
    Oct 5, 2026, 9:30 PM

    @hazelweakly "This was already horrifically difficult before AI and now it’s damn near impossible to get the point across when literally any trivial example passes muster"

    This is my day to day now.
    I have to explain every day why the vibe coded PR is not doing what it should, why it's only solving the problem on surface and the problem will surface again in XYZ situations.

    😮‍💨

    Preach, sending you hugs.

    The only thing that I can do that seems to work is to treat everyone as juniors (but this is an insult to juniors tbh) and start taking everyone by the hand even if they have lots of experience in this industry...

    "No, you should have asked the clanker to do this, and validate this, and not ask just for a fix, and not just copy/pasted your issue in there, etc."

    💬 0🔄 0⭐ 0
  • David Ppewnack@aus.social
    Oct 5, 2026, 10:36 PM

    @hazelweakly

    "Extraordinary claims require extraordinary evidence" attributed to Carl Sagan

    The burden of proof is on them. Albeit easier said than done.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • estelleestelle@mstdn.social
    Oct 6, 2026, 3:20 AM

    @hazelweakly windows is bloating and the ai targets wasted all the united states' ammo supply and the united states is still losing in iran and it can never be profitable and is demonstrably useless, but maybe we should keep going and see what happens

    💬 0🔄 0⭐ 0
  • Johan Sköldrhoot@mastodon.gamedev.place
    Oct 6, 2026, 3:37 AM

    @hazelweakly I was trying to explain to a friend that LLMs can't do math. They will spit out numbers that may or may not be correct, but it should be double checked since it doesn't actually understand those numbers. They kept arguing "but I've had it do math for me all day".

    What finally made it click for them was when I told them they should ask ChatGPT itself if it is able to do math, and if not, why not.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 2k_1152k_115@mastodon.social
    Oct 6, 2026, 8:57 AM

    @hazelweakly The deeper problem is algorithmic amplification. Accounts built on insults, outrage and personal attacks can spread faster than evidence-based analysis. Low-capital users often lack the tools to decode complex systems, so engagement becomes a substitute for empirical evidence.

    💬 0🔄 0⭐ 1