Skip to content
DevMeme
6119 of 7590
My AI-Powered Coding Prowess
AI ML Post #6710 · source on Telegram

My AI-Powered Coding Prowess

Description

A two-panel meme featuring simple, black-and-white stick-figure characters. In the top panel, a character is depicted in a classic thinking pose, hand on chin, with a look of concentration and confidence. The text next to it reads, 'Me writing great code.' The bottom panel reframes the scene: the same character is in the thinking pose, but a smaller, smiling character is physically holding the first character's hand to its chin, as if puppeteering the thoughtful gesture. Positioned between the two figures are the logos for OpenAI and Google's AI (appearing to be Gemini/Bard). This meme humorously deconstructs the idea of individual coding genius in the age of AI. It satirizes developers who might take full credit for their work, while heavily relying on AI coding assistants like ChatGPT or Gemini to generate the 'great code' they're so proud of. The joke is on the developer's inflated sense of self, oblivious to the AI acting as the ghostwriter

Categories

AI ML DeveloperProductivity Tooling IndustryTrends Hype CodeQuality

Common Tags

AIAssistants Chatgpt AITools AI DeveloperHumor Openai DeveloperMemes CodingHumor

Comments

53
Anonymous ★ Top Pick My new pair programmer is amazing. It never gets tired, types at the speed of light, and only occasionally hallucinates an entire API that doesn't exist
  1. Anonymous ★ Top Pick

    My new pair programmer is amazing. It never gets tired, types at the speed of light, and only occasionally hallucinates an entire API that doesn't exist

  2. Anonymous

    ChatGPT leans over my shoulder and within two suggestions converts my tidy functional pipeline into an AbstractFactoryBuilderFactory - proof its training data peaked in 2006

  3. Anonymous

    After 20 years of arguing about tabs vs spaces and defending our hand-crafted abstractions, we've finally found the ultimate code review partner: one that never questions our architectural decisions because it helped write them, creating the perfect echo chamber where every pull request is just us agreeing with ourselves at different confidence levels

  4. Anonymous

    Every senior engineer knows this feeling: you're in the zone, abstractions are elegant, patterns are pristine, and you're convinced you've just architected the Sistine Chapel of code. Then you step away for coffee, return with fresh eyes, and realize you've actually built a distributed monolith with circular dependencies that would make a graph theorist weep. The real wisdom isn't avoiding this trap - it's recognizing that the gap between 'writing great code' and 'having written great code' is exactly one code review, three production incidents, and a humbling conversation with your future self at 3 AM

  5. Anonymous

    About to model aggregates and invariants; the LLM autocompletes a 500‑line singleton utils.ts, swallows errors, and asks me to press Merge - apparently I’m the rubber duck now

  6. Anonymous

    Intending SOLID principles, shipping a dependency graph that defies Tarjan's algorithm

  7. Anonymous

    Pairing with ChatGPT: my “great code” becomes a confident, elegant snippet that aces the happy-path test, ignores idempotency, and deadlocks the moment reality introduces concurrency

  8. @SpYvy 1y

    Gemini 2.5 pro is better

    1. @deadgnom32 1y

      I am better

  9. Deleted Account 1y

    I am prompt developer

    1. _ 1y

      Does that means you develop apps promptly ?

      1. Deleted Account 1y

        Yes, still counting as apps maker tho.

  10. @mrYakov 1y

    Idk, for now llm struggle to even understand my ideas. Right now my pet project is some ecs like data structure that allow multithreading processing. And llm struggle to even understand how it work in general way, not to saying about a lot of corner cases with addition/deletion of components

    1. @RiedleroD 1y

      yea imo LLMs at the current level can't comprehend problems actual programmers are hired to solve

      1. @RiedleroD 1y

        or maybe I just suck at prompting. it's not a skill I care to hone

        1. @StalinLivesInUSA 1y

          Skill issue still 🪦

          1. @purplesyringa 1y

            if you're only working on problems LLMs can solve, that speaks about your skills more

            1. dev_meme 1y

              That's strange take LLM is just an instrument. It's how you use it what defines if it can solve some problem or not

              1. @purplesyringa 1y

                There's a ton of problems LLMs can't solve, or at least can't solve at a smaller cost than a human (including time spent). Riedler's (and mine as well) point was that a majority of tasks developers do, or at least good developers, are of this category

                1. dev_meme 1y

                  And this is what I’m talking about, because "solve" needs definition in that context, otherwise it depends solely on personal expectations

                  1. @purplesyringa 1y

                    Okay

                2. dev_meme 1y

                  Thanks to gemini 2.5 and it's context of 1M tokens I just solved a problem with reverse-enginering on which I gave up 2 years ago I had to load context for 700k tokens, full of different networks dumps, etc Its completelly different reality we live in right now

            2. @StalinLivesInUSA 1y

              Naah. I’m using LLMs to create problems to solve 💸

  11. @ashit_axar 1y

    It's funny cause I've tried some different LLMs to write a C++ OpenGL app with basic functionalities (such as zoom, pan, rotate). Neither of them could deliver a working one. It's even funnier cause every time i tried to debug and modify a portion, it's like you are talking to a new guy with a completely new mindset 😂😂

    1. @deadgnom32 1y

      yes. even trying some small things, like parts of the program, particular algorithms, which usually works out quite well. I noticed, they are oftentimes in different styles. while you can keep the leafs of the architecture a blackbox, it is a huge problem when one lets AI develop abstraction layers, because they need to be in consistent style, but after a while they look like you hired 50 different people and fired them immediately after they wrote 1 function.

      1. @deadgnom32 1y

        at some point it becomes a huge problem resulting in unmaintainable and inextensible codebase

        1. @deadgnom32 1y

          and by trying to feed LLM code to an LLM to fix it, you run into ML-autofagia, because of which it starts to produce even worse and worse results over time.

          1. @deadgnom32 1y

            it actually follows from the underlying math models, but some people don't know why, and some people don't tell why.

          2. @ashit_axar 1y

            Maybe LLMs perform good in other languages like python,... I've never tried. But they can give an overall example how your code structure should look like.

            1. @deadgnom32 1y

              no

            2. @deadgnom32 1y

              and yes for the second part

            3. @deadgnom32 1y

              I use LLM for inspiration, not for writing

              1. @deadgnom32 1y

                this works out well

              2. @deadgnom32 1y

                especially because llms produce errors uncommon for people, and they look okay at the first glance, and at the second and third oftentimes too, but people with a bit of proffecionality could never fail in such places, this results in very hardcore debugging. that's why I better read the code, and write it in a same manner as everything else by myself.

                1. @deadgnom32 1y

                  I work with lawyers, they tried sometimes to spare themselves some work, they can code what they need most of the time if it's simple enough. and recently they tried to use Claude to write code they could write in 2 days. and then I had to fix it for a week.

              3. @TheFloofyFloof 1y

                You could get the same results by paying a tweaker $10

                1. @deadgnom32 1y

                  here I get it for free and in plenty of contexts. I'm more than full stack.

    2. dev_meme 1y

      Which one did you used?

      1. @ashit_axar 1y

        Copilot, Qwen coder, Deepseek, Hermes, Mercury(this one feels different, it's a diffusion model, it's real fast)

    3. dev_meme 1y

      Btw, did you already found any devmeme's page where webgl is used?

      1. @ashit_axar 1y

        Not really, I'm not a web developer actually. Sometimes i build small Desktop apps for my own use. Mainly in C/C++ and WxWidgets for GUI.

        1. dev_meme 1y

          You don't need to be a webdev to know about basics of how internet/browsers work tho

  12. @H3R3T1C 1y

    The client will pay "only 10 USD" to make their software with LLMs (Vibe Coding).... then they need to pay me 10k to solve all the holes and crash of the software made by using the LLM

    1. @tema3210 1y

      A good thing for us, no?

      1. @H3R3T1C 1y

        💯

  13. @qwnick 1y

    Rage bait

  14. @purplesyringa 1y

    That's a funny gag, but that's it

  15. dev_meme 1y

    This was funny some time ago, but really, tokenizers are not complicated https://platform.openai.com/tokenizer

  16. dev_meme 1y

    Tho to be honest I'm not even sure if this screenshot is real 🌚

    1. @paranoidPhantom 1y

      Just took it in t3 chat

      1. @paranoidPhantom 1y

        It's 4o)

    2. @purplesyringa 1y

      Would you say that it could've been AI generated? 😈

      1. @sylfn 1y

        or it could just be devtools'd

More like this

More like this in AI ML →
AI ML AIAssistants Chatgpt
Usage Reset Joy, Then the Five-Hour Cap Returns

Usage Reset Joy, Then the Five-Hour Cap Returns

AI ML
LLM Shoggoth, Agent Harness, Your Code

LLM Shoggoth, Agent Harness, Your Code

AI ML
Agent Fixes the Bug By Growing a Body

Agent Fixes the Bug By Growing a Body

AI ML
Use J and K for navigation