Blake Richards

Is a universal brain decoder possible? Can we train a decoding system that easily transfers to new individuals/tasks? Check out our #NeurIPS2023 paper where we show that it’s possible to transfer from a large pretrained model to achieve SOTA 🧠! Link: poyo-brain.github.io 🧵

We’re excited to support @evadyer and @tyrell_turing as they combine different ways of measuring neural activity to better model how the brain works. They will explore the development of a general-purpose, multiscale, multimodal model of human brain activity that learns shared representations across invasive (e.g. intracranial EEG) and non-invasive (e.g. scalp EEG) recordings. The goal is to build a foundation for simulating, decoding, and interacting with brain dynamics in ways that advance both neuroscience and the development of more interpretable, brain-aligned AI systems.

Excited to share our #NeurIPS2025 work: NuCLR, a framework for learning neuron-level representations 🧠 These embeddings capture the biological identity of neurons and work out-of-the-box on new animals; no finetuning needed 💃 This offers some of the first evidence that large-scale neuroscience models can truly generalize across animals. Paper: arxiv.org/abs/2512.01199 Code: github.com/nerdslab/nuclr If you are at NeurIPS in San Diego, come find us at Poster Session 5 (11am-3pm PT, Exhibit Hall C,D,E, # 2107) 🎉 1/x 🧵

Excited to share our #NeurIPS2025 work: NuCLR, a framework for learning neuron-level representations 🧠 These embeddings capture the biological identity of neurons and work out-of-the-box on new animals; no finetuning needed 💃 This offers some of the first evidence that large-scale neuroscience models can truly generalize across animals. Paper: arxiv.org/abs/2512.01199 Code: github.com/nerdslab/nuclr If you are at NeurIPS in San Diego, come find us at Poster Session 5 (11am-3pm PT, Exhibit Hall C,D,E, # 2107) 🎉 1/x 🧵

Autoregressive language models learn to compress data by mapping sequences to high-dimensional representations and decoding one token at a time. The quality of compression, as defined by the ability to predict the next token given a prompt, progressively improves (as measured by negative log-likelihood) during training. We find that complexity of the representation manifold however, evolves non-mononitically in distinct phases across pretraining and post-training. Excited to share our #NeurIPS2025 📄 led by our amazing undergrad @melody_zixuan where we study the complexity dynamics of LLMs, and how distinct phases relate to specific behaviors. 🧵👇