Dosidicus
Raise digital squids whose brains grow & rewire themselves
by Rufus Pearce
FreemacOSWindowsLinux
- Upvotes
- 14
- Comments
- 0
- Impressions
- 263
this week
discussions
this week
About
Dosidicus is a cognitive sandbox presented as a digital pet squid. Each squid starts with a small, randomly wired brain and changes over time based on interaction. You can feed and stimulate the squid, then watch neuron activations and structure evolve as it learns. The project emphasizes transparency: the neural activity is visible and directly manipulable, contrasting with typical opaque AI models. Under the hood, it uses a custom simulation engine built in NumPy (not TensorFlow or PyTorch) and includes structural growth (neurogenesis), Hebbian-style rewiring, and a dual memory system. It supports a GUI app and a headless mode for running longer simulations without the interface, plus a save system to export and share a squid’s “cognitive history.” The repository also mentions modular components, configurable/custom brains, and a plugin system.
Problem & solution
Problem
AI learning is hard to observe and understand in black-box systems.
Solution
A visible digital pet squid NN that grows and rewires as you interact.
Who it's for
Learners and hobbyists exploring neural networks.
Main features
- Visible neuron activations
- Hebbian rewiring
- Neurogenesis growth
- Dual memory system
- Headless training mode
Use cases
- Raise and observe a learning squid
- Run headless training simulations
Comments
0 shownShare your thoughts about this tool.
Sign in to leave a comment.
No comments yet. Be the first to leave one.