Hi everyone, I’m a developer currently starting an indie game company, and I’ve been exploring open source AI tools that can help me build smart, adaptive systems for games and creative tools. Like many, I’ve used LLMs and found them incredibly powerful—but I also believe we’re going to hit diminishing returns if we keep scaling in the same direction (more GPUs, more data, etc.). So I’ve been searching for alternatives that more closely model how intelligence works in the brain—and that’s how I stumbled across The Thousand Brain Project, thanks to a ChatGPT suggestion.
After watching the videos and diving into the docs, I really like the long-term vision behind TBP and the Monty SDK. It feels like the kind of foundational project that could power a new wave of AI systems—and I’d love to contribute as I go deeper. I’m still ramping up, but already I’ve got a few experimental ideas, like designing a “curved finger” sensor that mimics how human fingers perceive by exclusion—how areas not in contact with objects still provide useful info. Another idea is rendering the finger in 3D so it’s visible to the agent’s eye, helping the model learn to distinguish between itself and the external world—potentially useful for robots manipulating objects.
Really excited to learn more and contribute to this community!
That’s great to hear, those both sound like really interesting ideas. The idea of having a 3D model of the finger so that the distant agent can sense it is particularly interesting, and could fit in well with our objective of a having a cross-modal policy that combines these two agents.
Once you’re feeling up for it, a great starting point would be to write out an RFC (Request for Comment), which you can read more about here. If you’re unsure about the process, you can take a look at some of the previous RFCs that have been written (these will appear under closed pull requests in tbp.monty), or ask clarifications here.
Thanks again for your involvement and looking forward to hearing more soon.
Thanks @nleadholm for the warm welcome! I’m really looking forward to contributing.
I tried running through the “Getting Started” steps from the docs, but it looks like they don’t currently work on Windows. I use Ubuntu via WSL, so I figured it might just work—but I ran into some conda-related errors. I’ll dig into it a bit more and see if I can figure it out. Who knows, maybe my first PR will be getting it working on WSL.
Once I get up to speed with the codebase, I’ll definitely start drafting some RFCs. Excited to be part of the community!
Hi @AdamLD welcome!
Just to mention, there has been some discussion around using Monty on windows in this Discourse thread and the corresponding issue on GitHub. Generally Monty should work on Windows but as @tslominski mentions on the Issue you will not be able to use the Habitat Simulator (which we use in our benchmark experiments and some unit tests) since this simulator doesn’t run on Windows. I think you have to manually remove this dependency from the environment yml file as described here Windows OS not supported? · Issue #52 · thousandbrainsproject/tbp.monty · GitHub
Basically I run Ubuntu inside my Windows. There is a mounted shared folder in between the systems. So I’m able to use the code editor in Windows, but the bash shell is in the Ubuntu system, and able to run all the commands without any issues.
The problem I had was conda & pip related. Basically pip was silently upgrading my pytorch version, which was causing the pytest command to fail the tests.
I’ve opened an issue in github with more details about this.
