Why it matters, why it resists measurement, and how to govern automated development
Version 1.26 · Updated 22 August 2026 · Evidence cut-off 22 August 2026 · Links checked 22 August 2026
Written and researched by OpenAI Codex, under the direction and editorial guidance of Peter Roelants
What they are, how they work, and what it takes to run one
Version 3.7 · 23 August 2026
Written and researched by agents such as Codex, under the direction and editorial guidance of Peter Roelants
Working definition. A dark software factory is a development system in which people specify intent, constraints, and acceptable behavior while software agents produce and validate the implementation. In the strict form, people neither write nor routinely review the generated code; they operate the factory that does.
| from transformers import AutoTokenizer, T5ForConditionalGeneration | |
| # Model Init | |
| n_gpu = 8 | |
| tokenizer = AutoTokenizer.from_pretrained("google/flan-ul2") | |
| model = T5ForConditionalGeneration.from_pretrained("google/flan-ul2") | |
| heads_per_gpu = len(model.encoder.block) // n_gpu | |
| device_map = { | |
| gpu: list( | |
| range( |
| """ Trains an agent with (stochastic) Policy Gradients on Pong. Uses OpenAI Gym. """ | |
| import numpy as np | |
| import cPickle as pickle | |
| import gym | |
| # hyperparameters | |
| H = 200 # number of hidden layer neurons | |
| batch_size = 10 # every how many episodes to do a param update? | |
| learning_rate = 1e-4 | |
| gamma = 0.99 # discount factor for reward |
Code for Keras plays catch blog post
python qlearn.py
| """ | |
| Minimal character-level Vanilla RNN model. Written by Andrej Karpathy (@karpathy) | |
| BSD License | |
| """ | |
| import numpy as np | |
| # data I/O | |
| data = open('input.txt', 'r').read() # should be simple plain text file | |
| chars = list(set(data)) | |
| data_size, vocab_size = len(data), len(chars) |
| """ | |
| A deep neural network with or w/o dropout in one file. | |
| License: Do What The Fuck You Want to Public License http://www.wtfpl.net/ | |
| """ | |
| import numpy, theano, sys, math | |
| from theano import tensor as T | |
| from theano import shared | |
| from theano.tensor.shared_randomstreams import RandomStreams |
| #!/bin/bash | |
| # Pylearn2 setup script for Amazon Linux AMI with NVIDIA GRID GPU Driver. | |
| # http://goo.gl/3KeXXW | |
| # not tested | |
| sudo yum update -y | |
| sudo yum install -y emacs tmux python-pip | |
| sudo yum install -y python-devel git blas-devel lapack-devel |
| # (C) Kyle Kastner, June 2014 | |
| # License: BSD 3 clause | |
| import scipy.stats as st | |
| import numpy as np | |
| class gmmhmm: | |
| #This class converted with modifications from https://code.google.com/p/hmm-speech-recognition/source/browse/Word.m | |
| def __init__(self, n_states): | |
| self.n_states = n_states |
