AI ML
Post #7493, on Nov 28, 2025 in TG
Hide the Pain Harold Lost His Job to a Sigmoid Activation Function
Description
The iconic 'Hide the Pain Harold' meme featuring an elderly man with a forced smile and visibly pained eyes. The top text reads 'I LOST MY JOB TO' and the bottom shows the mathematical formula for a neural network's sigmoid activation function: sigma(b + sum from i=1 to n of w_i * x_i), which is the standard formula for a single neuron's output -- the sigmoid function applied to a weighted sum of inputs plus bias. An imgflip watermark is visible. The meme captures the existential dread of being replaced not by some sophisticated AI system, but by what is essentially the simplest possible neural network unit -- a single perceptron with a sigmoid activation
Related Memes
Use J and K for navigation
Comments
32Comment deleted
Getting replaced by a sigmoid is humiliating enough, but wait until you learn it was actually a ReLU -- they didn't even need the fancy squishing function
After writing this, the editor lost their job to AI Comment deleted
Human compilers watching this meme and remembering old days Comment deleted
Probably deserved it Comment deleted
these organic llms think they are superior Comment deleted
I'm not very successful in mathematics but can a robot write working code? Comment deleted
Bruv check out Claude Code Comment deleted
It's ok for spotting issues, but not good at fixing them or doing anything complex from scratch Comment deleted
Idk what language or Framework you use, but it works nicely with PHP and JS Comment deleted
Ok, let's be real for a second (nerdy boring stuff) What it can do: - help you brainstorm - potentially replace a junior dev - get simple solutions quickly (or at least a direction) - follow the repo code style (at least it tries to) What it can't do: - find simple solutions for complicated problems without bloating your codebase - write useful unit/integration tests - prepare good quality code (readability, optimization, security) I work with Python and even if you magically manage to make it generate a piece of code that's functional, it will still fail most of static checks, violate peps and so on and so forth, essentially it will spit out senior-level concepts in a junior-level implementation (simplification, but you get the point). And this is only a part of it. Considering you are a senior and considering the amount of time you spend for writing a good prompt and re-iterating, you might as well spend a similar amount of time to implement the solution yourself, which will be a lot more suitable and "better", not even mentioning monetary loss for running your prompts. Comment deleted
You say what it can’t do You don’t say what you tried to use/tested and what was your setup This is dead-ended conversation Comment deleted
Fair enough, let's consider one situation: The task: Implement hybrid search (full text + semantic), can't disclose implementation details. It had to use a specific library for semantic search and find other suitable ones if necessary. The setup: Postgres + vector plugin, python 3.13, poetry. Prepared a detailed prompt, thoroughly explaining the expectations, refined it 3 times. Outcome: Non-functional class, that didn't even seem to be close to what I described in the prompt. Could be a skill issue ofc, but again, is it worth investing time into preparing a prompt and iterating compared to investing time into implementing it yourself? I wouldn't say so for complex tasks Comment deleted
But you still didn’t mention LLM that was used and via which tool you used this LLM Comment deleted
does it matter? regardless of their answer you're going to say "well try another LLM" Comment deleted
> is it worth investing time into preparing a prompt and iterating compared to investing time into implementing it yourself? In actual work setting you better prepare and discuss Solution Design with your team(-s) before you proceed with implementation 🌚 Comment deleted
Idk where you work but I haven't seen anyone caring, even in big tech. All they say is "if you want to use ai use what we allow you to use, it's up to you if you want to use it". At least from my experience. Comment deleted
same thing in insurance companies, they are more interested about projects rather then devs doing vibe coding Comment deleted
It wasnt a comment about AI, Solution Design in what you create for the task that you have been assigned with Comment deleted
in the end it is same pigeon hole issue like with every generative AI Comment deleted
I brainstormed with chatgpt for 30 minutes trying to debug with code, perfectly fine code, error was in environment Comment deleted
it's not good at it either Comment deleted
Idk maybe your prompts are wrong or the data set is irrelevant, it works for me Comment deleted
Good but I still gotta fix it often Comment deleted
Bro, it's scientifically proved, if you give enough time doing random keyboard hits, even a chimp will eventually write any program ever existed... Comment deleted
That I don't doubt :3 Comment deleted
the warm bias in AI-generated bullshit needs to be studied Comment deleted
you mean piss filter? Comment deleted
this is because of boom of pfp's in Hayao Miyazaki's style generated by chatgpt, so during training it favoured scenes form Miyazaki's work, wich are mostly warm colors. At least that what I remember Comment deleted
it feels strange why high usage of prompting it to fetch miyazaki weights would affect the training process at all Comment deleted
OpenAI literally used this in their ads, so I think they might've favoyred some images over other Comment deleted
Also has (A LOT) to do with old photos being yellowed Comment deleted
All hail to our soon-to-be AI overlords, or overladies. Comment deleted