Login
You're viewing the aleph.land public feed.
  • Oct 29, 2026, 5:48 PM

    As I try furiously to compete with AI - both in code and in person, I'm realizing Claude is way much more capable than I at navigating and debugging unknown code.

    It inspects every nook and cranny of the codebase, finding configuration bugs, weird edge cases, etc.

    How the hell does it do that?

    I still think it's unethical to depend on giant data centers to do this. But... How??? It's a really capable tool, and I feel... threatened by it.

    #ai

    💬 3🔄 0⭐ 0

Replies

  • Oct 29, 2026, 5:57 PM

    If it uses software to check for those bugs, well, that'd be one thing, I can perfectly understand tree traversal, the traveling salesman and those hard-NP problems we read about in college, but as it is right now? I really don't know if it does.

    If it's just an LLM inferring answers from its memory matrices... and it STILL finds those bugs... hell, how am I ever going to compete with that? 😟

    💬 1🔄 0⭐ 0
  • Oct 29, 2026, 6:09 PM

    @yuki2501 Well that's the thing right? You don't compete with tools, you use them to do your job better. It's my hope that all these people that think it's a genie will fall to the wayside. There may be a bit of a paradigm shift, but I still believe the cream rises to the top.

    It's either that or we're doomed. Kind of a tossup right now

    💬 1🔄 0⭐ 0
  • Oct 30, 2026, 4:35 PM

    @jackgangi No no, I think you're right. Just yesterday I had a horrible hiccup with AI tools at my job.

    I was using Claude to fix some bugs after dealing with a merge and then I had the brilliant idea 🙄 to let Copilot do a code review.

    Copilot fixed the bugs but introduced a really nasty bug that took Claude over an hour to fix.

    At the end, the PR doubled in size.

    I felt like going to a deathbot competition, but instead of fighting each other, the bots were shredding my code. 😱

    💬 1🔄 0⭐ 0
  • Oct 30, 2026, 4:37 PM

    @jackgangi As I researched more on AI models capabilities, it dawned on me: It IS, effectively, just math. A really convoluted way of doing stuff, based not on the actual code structure, but on the code's TEXT.

    This is why they need billions of parameters to function properly. Because otherwise they can't see shit.

    💬 1🔄 0⭐ 0
  • Oct 30, 2026, 4:44 PM

    @jackgangi And I'm thinking... Hey, surely there's an ALGORITHMIC way of doing this better, isn't there?

    I'm going to try using approaches like abstract syntax trees and code execution flows to see how good AI actually is compared to optimized algorithms.

    💬 0🔄 0⭐ 0
  • Oct 29, 2026, 6:33 PM

    @yuki2501 dont anthropomorphize the statistical blender trained on the theft of everyones git commits.

    software isnt insurmountable, debugging isnt magic, its just a process.

    just practice your own skills and decision trees, like any good chess player, or writer would.

    💬 1🔄 0⭐ 0
  • Oct 29, 2026, 6:59 PM

    @discatte I'm not anthropomorphizing it, but I don't think it's correct to call it software.

    It's its own thing, it's neither alive nor software. So... deepware, I guess? 🤔

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Oct 29, 2026, 10:20 PM

    @yuki2501 I think well hand-written code is gonna be that, an artisanal craft. We are experiencing what the carpenters and blacksmiths must've felt during the industrial revolution and even more when synthetic materials replaced hand-carved wood or forged iron. They are still here but for a more specialized and expensive craft.
    You must decide what kind of craft you wanna do and also depending on what the market will ask for.

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0