Show HN: Fly Language Model (flm.kuber.studio)

A new experimental language model, dubbed the Fly Language Model (FLM), has been developed to simulate neural dynamics inspired by the connectome of a fruit fly's brain. Unlike traditional models that rely on pretrained transformers, FLM was trained from scratch using a derived recurrent network architecture based on anatomical wiring. This approach enables researchers to analyze how specific changes in neural structure and connectivity influence language processing and model behavior, isolating factors such as memory dynamics and computational changes. The significance of FLM lies in its potential to enhance our understanding of neural computation and memory in language models, providing a novel framework for exploring how biological principles can inform AI and machine learning. The model's architecture operates with fixed directed communication graphs and multiscale memory systems, which contrast with the adaptive, attention-driven systems seen in prevailing transformer models. By incorporating controlled experiments that manipulate neuronal communication and state responses, FLM opens avenues for investigating the transferability of learned behaviors and the role of specific neural configurations in cognitive tasks, thereby contributing valuable insights to the intersection of neuroscience and artificial intelligence.