Thousand Brains Project
@thousandbrains.org
Advancing AI & robotics by reverse engineering the neocortex. Leveraging sensorimotor learning, structured reference frames, & cortical modularity. Open-source research backed by Jeff Hawkins & Gates Foundation. Explore thousandbrains.org
📊 What’s next for the Monty platform? @tristanls.bsky.social explores how we’re separating Monty from experiment infrastructure and simulated environments, and building a telemetry pipeline that emits standardized data as events happen. Watch the full video here: youtu.be/G9CqGYArJLQ
🎮 How would an AI solve a problem without using language or external knowledge? @vivianeclay.bsky.social explores ARC-AGI-3, a benchmark test for artificial intelligence, and how it could become a practical testing environment for new feature prototypes in Monty.
🧠 How does the brain build, update, and switch between maps of the world? @vivianeclay.bsky.social dives deep into grid cell literature, exploring how they anchor based on landmarks, remap as environments change, switch between reference frames, and operate across multiple spatial scales.
Benji shares a big thank you to the community for their valuable contributions in July 2026. 🫶🏼 Watch the full video here: youtu.be/wL0GuAWlXWM
🧠Ready for a whirlwind of neuroscience research? @tristanls.bsky.social recaps research that stood out to him from the recent FENS conference. Topics include new findings on how the brain maps space, prepares for action, corrects errors, makes predictions, and forms memories.
How does Monty perform in the real world? Zachary Danzig shares his undergraduate thesis, which brings Monty to real robotic hardware. His proof of concept project combined stereo vision, depth sensing and a robotic arm for object recognition. Watch the full video here: youtu.be/3wgHNqkRekM
🔍 What open questions remain for the Thousand Brains Theory? @vivianeclay.bsky.social guides a wide-ranging discussion on attention, scale, unsupervised learning, model representations, and the role of goals and actions. Watch the full video here: youtu.be/AGp6zn_IO-s
🎯 How should attention guide perception? @cortical-canonical.bsky.social and @canoetrip.bsky.social explore attention as an area-based constraint to guide Monty’s resources, and how we can use salience maps and flood-fill algorithms to find regions of interest.
📍How do reference frames work in the brain? How could they work in the neocortex? Jeff and @hojaeklee.bsky.social discuss papers on grid cells, place cells, path integration, and overlapping grid modules, and lead a high-level exploratory discussion.
Could overlapping grid cell modules help explain unresolved questions about reference frames in the thousand brains theory? Jeff sparked a discussion about how the neocortex might represent unique locations. No conclusions just yet. Watch the full video here: www.youtube.com/watch?v=469n...
🧠 @canoetrip.bsky.social shows how visual salience can guide Monty’s sensors to move to the most informative features. The result: sparser models, fewer movements, and no loss in accuracy. Watch the full video here: www.youtube.com/watch?v=AE-Z...
@vivianeclay.bsky.social reviews our progress in Q2 2026 and goes over our plans for the next quarter. Watch the full video here: youtu.be/MtrvhYQ6VFw
@vivianeclay.bsky.social shares her gratitude to the community for their valuable contributions in June 2026. 🫶🏼 Watch the full video here: www.youtube.com/watch?v=f-lx...
🤖 How does Monty know where to look next? 👀 @vivianeclay.bsky.social and @cortical-canonical.bsky.social discuss attentional regions, how attention relates to policies, gating sensory input and voting, and whether prediction error can help guide Monty’s behavior.
🧠 What changes when Monty goes from one column to multiple columns? Jeff presents key challenges, then brainstorms around how information can be shared during learning and inference. Watch the full video here: www.youtube.com/watch?v=9r3n...
🦾 In this meeting, @vivianeclay.bsky.social goes over a high-level overview of Monty’s capabilities (current and future) as well as unresolved theoretical questions. Watch the full video here: youtu.be/vc7O4Ludbcw
💡 @vivianeclay.bsky.social gives a summary of a research meeting where we do a deeper dive into trajectory memory for behavior models. Watch the full video here: www.youtube.com/watch?v=y3zB...
🧠Jeff gives a summary of a research meeting where we discuss trajectory memory for behavior models and shared learning. Watch the full video here: www.youtube.com/watch?v=Fcfv...
🚀 A big step forward for Monty! Hojae introduces the new 2D sensor module to deal with 3D distortions. This enables Monty to learn a logo on one surface and now recognize it on any other surface! Watch the full video here: www.youtube.com/watch?v=Gi96...
A big thank you from @vivianeclay.bsky.social to the community for their valuable contributions in May 2026. 🫶🏼 Watch the full video here: www.youtube.com/watch?v=iSUq...
🧠 Jeff gives a summary of a research meeting where we discuss open theory questions. Watch the full video here: www.youtube.com/watch?v=tBu2...
📖 Our paper “Thousand-Brain Systems: Sensorimotor Intelligence for Rapid, Robust Learning and Inference” is now published in Neural Computation! Read more about how Monty works, its capabilities, and advantages over transformers: direct.mit.edu/neco/article...
Thousand-Brains Systems: Sensorimotor Intelligence for Rapid, Robust Learning and Inference
Abstract. Current AI systems achieve impressive performance on many tasks, yet they lack core attributes of biological intelligence, including rapid, continual learning, representations grounded in se...
direct.mit.edu
Thousand Brains Project@thousandbrains.org · last yr./1 🚨 New Paper Drop 🚨 We’ve released “Thousand-Brains Systems: Sensorimotor Intelligence for Rapid, Robust Learning & Inference” — the first working implementation of a thousand-brains system, code-named Monty. 👇 Dive into the full thread 🧵 arxiv.org/abs/2507.04494
✨ @vivianeclay.bsky.social gives a summary of a research meeting where we discuss state in morphology models. Watch the full video here: www.youtube.com/watch?v=z3VY...
📖 Our paper “Thousand-Brain Systems: Sensorimotor Intelligence for Rapid, Robust Learning and Inference” is now published in Neural Computation! Read more about how Monty works, its capabilities, and advantages over transformers: direct.mit.edu/neco/article...
Thousand-Brains Systems: Sensorimotor Intelligence for Rapid, Robust Learning and Inference
Abstract. Current AI systems achieve impressive performance on many tasks, yet they lack core attributes of biological intelligence, including rapid, continual learning, representations grounded in se...
direct.mit.edu
🤌 @cortical-canonical.bsky.social gives a summary of a research meeting where we discuss the Milan Week takeaways and ideas around object state and behaviors. Watch the full video here: youtu.be/1vNRku9BZHM
📝 @ramymounir.com gives a summary of a research meeting where we discussed voting and shared learning using temporal memory and spatial pooler-like mechanisms. Watch the full video here: www.youtube.com/watch?v=9vB2...
🧠 @vivianeclay.bsky.social recaps key takeaways from a recent focus week on how the brain could learn to accomplish goals. Watch the full video here: www.youtube.com/watch?v=TRgq...
🧠 Jeff reviews literature on the hippocampal complex and presents new ideas from a recent focus week around learning temporary models. Watch the full video here: youtu.be/zCMCCwJEmmc
✨ @vivianeclay.bsky.social gives a summary of a research meeting where we discuss new insights around deformations, location sharing, and affordances. Watch the full video here: www.youtube.com/watch?v=cJLH...
☕ @cortical-canonical.bsky.social summarizes a recent research meeting in which we discussed top-down connections and other topics related to modeling compositional objects. We also touched on concepts like attention and leveraging different model types across columns.