Fly Language Model: Fly Language Model
Train a language model built from fruit fly connectome. Watch neural activity change with each token. Run controlled experiments: silence neurons, inspect state, measure memory contributions.
Compiled automatically from public sources — the company's own posts, launch material and reporting, all linked at the bottom. The “what using it looks like” section is a walkthrough built from public documentation, not a first-hand review. If this is your company, claim the page and correct anything wrong.
Train a language model built from fruit fly connectome. Watch neural activity change with each token. Run controlled experiments: silence neurons, inspect state, measure memory contributions.
What it does
Browser-based research interface to train and inspect a language model using a recurrent network derived from fruit fly connectome. Supports local learning, neural visualization, and controlled interventions.
Who it's for
Researchers interested in neural computation, interpretability, and understanding how memory and recurrence contribute to language modeling.
Where it fits
- Train a model on text and inspect neural states
- Silence neurons to measure their contribution
- Study memory timescales in language prediction
The short read
- Most language models are black boxes inside a transformer.
- Replays same input before and after training; controlled interventions measure contributions.
The idea
What if you could see what changes when a language model learns? This connectome-derived architecture makes the computation visible.