Contract 3 can't be done.

Contracts 1 and 2 go fine. One Neuron and a couple of hundred Passes, and your Network tells light from dark and left from right.

Contract 3 asks whether two pixels are the same or different. The Network stalls, and more training doesn't move it.

In 1969 a book proved it never would, and interest in neural networks collapsed for years.

What this is

An incremental game where the thing you upgrade is a real neural network. It trains in your browser while you play, on real images, with real backpropagation. So it fails the way real ones do: it gets stuck, it memorises its Training Samples, and it thrashes when you push the Learning speed too far.

The way past Contract 3 is the one the field found in 1986. It arrives with a new computer: the Lab starts on a 1958 terminal and moves to a 1986 workstation.

Quick answers

  • Length: Chapter 1 is about 40 minutes, six Contracts.
  • Active or idle: you click each Pass at first. Autorun, Labellers and Deployments take over.
  • While you're away: Deployments keep earning, at half rate, for up to two hours.
  • Prestige: Chapter 1 ends at the first one, Publish. The loop is the full game.
  • Saves: in your browser, with export and import in Settings.
  • Price: free. Nothing to buy, and no ads.
  • Runs on: a desktop browser. It isn't made for phones.

The full game

Chapter 1 is the first of four. The full game, about five hours, is coming to Steam.

Where did you get stuck?

Tell us in the Lab's Discord, games-q.com/discord. It decides what gets fixed first.

Published 2 days ago
StatusIn development
PlatformsHTML5
Rating
Rated 3.0 out of 5 stars
(3 total ratings)
AuthorGames?
GenreSimulation, Educational
Made withPixiJS
Tags1-bit, Idle, Incremental, machine-learning, neural-network, Pixel Art, Retro, science, Singleplayer
Average sessionAbout a half-hour
LanguagesEnglish
InputsKeyboard, Mouse
AI DisclosureAI Assisted, Code, Text

Comments

Log in with itch.io to leave a comment.

(+2)

I'm assuming most of the text on this page and in the game is AI generated. It hurts to read. It's like reading texts between two people (who have shared but hidden context) and having to do detective work to build the missing context. So the helper (PIP?) isn't being instructive, it talks like it's reminding you of something you are already supposed to know, and it's pretty irritating.

Hello, you’re right about the game text. We are rewriting Pip’s lines and making sure the text is easier to follow. One of the features we are working on is contract introductions - that should solve the issue of Pip talking about something you should already know.

Thank you for the feedback! It means a lot to us