@GoodfireAI: Neural networks might speak English, but they think in shapes. Understanding their rich *neural geometry* is key to und…
Summary
Goodfire AI announces a new research agenda focused on neural geometry to improve the understanding, debugging, and control of neural networks.
View Cached Full Text
Cached at: 05/08/26, 10:46 AM
Neural networks might speak English, but they think in shapes.
Understanding their rich neural geometry is key to understanding how they work – and to debugging and controlling them with precision.
Starting today, we’re releasing a series of posts on this research agenda. 🧵 https://t.co/CE3Xw7kFGV
Similar Articles
@GoodfireAI: Neural networks do math by rotating shapes. We found a shape-rotating calculator hidden inside an LLM – and it’s used f…
GoodfireAI found that neural networks perform math by rotating shapes, uncovering a shape-rotating calculator inside an LLM that is used for more than just math.
@FinanceYF5: Neural Networks Speak English, But They Think in "Shapes" 1/ Neural Networks Don't Think in Words They appear to speak English on the surface, but internally they may organize information in geometric space: curves, loops, surfaces, manifolds. Understanding neural geometry may be the key to understanding, debugging, and controlling models.
Neural networks appear to speak English on the surface, but internally organize information in geometric space (curves, loops, surfaces, manifolds). Understanding "neural geometry" may be the key to understanding, debugging, and controlling models.
We removed an LM's ability to speak German (3 minute read)
GoodfireAI releases a research agenda on understanding neural geometry in language models, demonstrating the ability to precisely control a model's capabilities, such as removing its ability to speak German.
@k_solidified_: https://arxiv.org/abs/2106.10165 All of humanity should read this
This book develops an effective theory for deep neural networks, showing that their predictions are nearly-Gaussian and governed by the depth-to-width ratio, and introduces representation group flow to analyze signal propagation and learning dynamics.
Can SAEs Capture Neural Geometry? (6 minute read)
This article explores how sparse autoencoders (SAEs) can capture curved neural geometry, revealing three distinct ways SAE features represent manifolds, and presents an unsupervised pipeline to uncover geometric structure in neural representations.