Niels Leadholm
@cortical-canonical.bsky.social
Research lead @thousandbrains.org (AI/ML)
Prev. RS @numentaofficial.bsky.social + PhD in neuro-inspired AI at @ox.ac.uk
Working on sensorimotor machine intelligence
Prev. RS @numentaofficial.bsky.social + PhD in neuro-inspired AI at @ox.ac.uk
Working on sensorimotor machine intelligence
... both about next-generation AI, as well as top-tier software development.
Flexible/remote work available.
thousandbrains.org/company/care...
(2/2)
Flexible/remote work available.
thousandbrains.org/company/care...
(2/2)
Open-Source Software Engineer
As a software engineer on the Thousand Brains Project, you will shape our open-source code base to make it accessible to a large research community. Additionally, you will write code and maintain infr...
thousandbrains.org
August 4, 2025 at 10:26 AM
... both about next-generation AI, as well as top-tier software development.
Flexible/remote work available.
thousandbrains.org/company/care...
(2/2)
Flexible/remote work available.
thousandbrains.org/company/care...
(2/2)
In case you missed it, check out our ML-focused paper on the current capabilities of Monty:
- Robust object reps w/ symmetry
- Model-free + model-based policies
- Rapid inference + learning
- Continual learning
- FLOPs efficient (8+ orders of magnitude)
bsky.app/profile/cort...
- Robust object reps w/ symmetry
- Model-free + model-based policies
- Rapid inference + learning
- Continual learning
- FLOPs efficient (8+ orders of magnitude)
bsky.app/profile/cort...
Ok this is one I've been really excited to share - our pre-print demonstrating the capabilities of thousand-brain systems is now out 🧠🚀 It gives a great overview of why we believe the approach we're taking is so promising. Rapid and continual learning leading to rapid and...
1/
This week, we’re releasing two milestone papers: one shows the amazing capabilities of thousand-brains systems and their benefits over deep learning, the other proposes a new theory of long-range connections in the neocortex. Years of work led to this.
This week, we’re releasing two milestone papers: one shows the amazing capabilities of thousand-brains systems and their benefits over deep learning, the other proposes a new theory of long-range connections in the neocortex. Years of work led to this.
July 9, 2025 at 6:20 PM
In case you missed it, check out our ML-focused paper on the current capabilities of Monty:
- Robust object reps w/ symmetry
- Model-free + model-based policies
- Rapid inference + learning
- Continual learning
- FLOPs efficient (8+ orders of magnitude)
bsky.app/profile/cort...
- Robust object reps w/ symmetry
- Model-free + model-based policies
- Rapid inference + learning
- Continual learning
- FLOPs efficient (8+ orders of magnitude)
bsky.app/profile/cort...
...robust inference - all essentially for free in a system inspired by the architecture of the mammalian neocortex. If you've had the growing feeling that training transformers on an internet worth of data might not be the path to true intelligence, then this is probably for you.
July 8, 2025 at 2:03 PM
...robust inference - all essentially for free in a system inspired by the architecture of the mammalian neocortex. If you've had the growing feeling that training transformers on an internet worth of data might not be the path to true intelligence, then this is probably for you.
global output of a transformer. Together with direct sensory inputs (not isomorphic to skip connections in DNNs), this means that the cortex (and Monty) is made up of many direct sensorimotor loops, while also having a hierarchical arrangement (see also www.nature.com/articles/s41...).
December 12, 2024 at 6:18 PM
global output of a transformer. Together with direct sensory inputs (not isomorphic to skip connections in DNNs), this means that the cortex (and Monty) is made up of many direct sensorimotor loops, while also having a hierarchical arrangement (see also www.nature.com/articles/s41...).
One point that I didn't mention because we weren't as focused on policies in March 2023 - every cortical column in the brain, and therefore every learning module in Monty, has a direct projection to sub-cortical motor-type regions. This is very different from the final,
(1/2)
(1/2)
How deep is the brain? The shallow brain hypothesis - Nature Reviews Neuroscience
Architectures in neural networks commonly assume that inference is hierarchical. In this Perspective, Suzuki et al. present the shallow brain hypothesis, a neural processing mechanism based on ne...
www.nature.com
December 12, 2024 at 6:18 PM
One point that I didn't mention because we weren't as focused on policies in March 2023 - every cortical column in the brain, and therefore every learning module in Monty, has a direct projection to sub-cortical motor-type regions. This is very different from the final,
(1/2)
(1/2)