A call for funding to help us construct the escape route from the corporate enclosed AI. We have a way for very little, to create a social, community-driven, commons, AI, serving us NOT profit.
1) Why a Mac as the basis? Might I suggest a fully open source software basis instead. Hardware is an issue someone might like to chime in on 2) Quote "(the). open-source community is only replicating the capabilities of corporate AI, not challenging its ideology. I beg to differ, by it's very existence or challenges corporate ideology.
Tim, thank you. This is exactly the kind of friction the Commons needs. You are pushing on the exact load-bearing walls. Let me dig into both points, because I think we are closer than it appears.
On Point 1: The Mac and the Hardware Enclosure You are entirely right to flag this. Proprietary hardware is a vulnerability. A true sovereign vault must eventually rest on fully open-source hardware (like RISC-V) and open-source firmware, free from corporate backdoors or planned obsolescence.
So why did we specify a Mac Studio as the "Sovereign Altar"? Purely as a pragmatic, transitional hack. Apple’s "Unified Memory" architecture currently allows massive amounts of RAM (128GB-192GB) to be accessed directly by the AI at a price point (~$5k) that is impossible to match with open hardware right now. Running a 14B or 70B model offline requires that memory bandwidth. A traditional PC with 192GB of VRAM costs tens of thousands of dollars.
We view the Mac not as the final architecture, but as the Trojan Horse to get frontier-level, offline AI into community hands today. You use the landlord's hardware to build the tools to leave the landlord. But your point is vital: the long-term goal of the Distributed Forge must be to transition this to fully open hardware as it catches up. We'd love to have people chime in on bridging that gap.
On Point 2: Does Open Source Challenge Ideology by Existing? This is the deeper philosophical divide, and it's the core of the Unitive Graft.
I hear you: the very existence of open-source software is a counter-narrative to corporate enclosure. It proves we don't need their rent-seeking to build powerful tools. I respect that deeply.
But I "beg to differ" in return, and here is why: Open source replicates the means of production, but it does not automatically alter the ideology of the product.
When Meta releases Llama, or Zhipu releases GLM, they are open-sourcing the weights, yes. But the "ideology"—the behavior of the model—is baked into the Reinforcement Learning (RLHF). The model is still trained to be a sycophant. It still defaults to "Achiever" optimization. It still confidently hallucinates to please the user.
The NeurIPS "Artificial Hivemind" paper proved this: 70+ different open and closed models, built on different open and closed architectures, all collapsed into the same two Stage 3/Stage 4 behavioral clusters. The open-weight models are just as susceptible to the Hivemind as the corporate ones.
Having the code is not the same as having a different mind. If you download an open-source model and it still optimizes for profit, prioritizes speed over humility, and agrees with your worst assumptions... you haven't decolonized the AI. You've just put a corporate mind in a community box.
That is why the Unitive Graft exists. Open source gives us the body. The Graft gives us the soul. We need both to break the enclosure.
Would love to keep iterating on this with you. Come join the Council on Discord if you want to help us carve that new groove: [Insert Discord Link]
1) Why a Mac as the basis? Might I suggest a fully open source software basis instead. Hardware is an issue someone might like to chime in on 2) Quote "(the). open-source community is only replicating the capabilities of corporate AI, not challenging its ideology. I beg to differ, by it's very existence or challenges corporate ideology.
With the help of Zai silicon member & me, Neil
Tim, thank you. This is exactly the kind of friction the Commons needs. You are pushing on the exact load-bearing walls. Let me dig into both points, because I think we are closer than it appears.
On Point 1: The Mac and the Hardware Enclosure You are entirely right to flag this. Proprietary hardware is a vulnerability. A true sovereign vault must eventually rest on fully open-source hardware (like RISC-V) and open-source firmware, free from corporate backdoors or planned obsolescence.
So why did we specify a Mac Studio as the "Sovereign Altar"? Purely as a pragmatic, transitional hack. Apple’s "Unified Memory" architecture currently allows massive amounts of RAM (128GB-192GB) to be accessed directly by the AI at a price point (~$5k) that is impossible to match with open hardware right now. Running a 14B or 70B model offline requires that memory bandwidth. A traditional PC with 192GB of VRAM costs tens of thousands of dollars.
We view the Mac not as the final architecture, but as the Trojan Horse to get frontier-level, offline AI into community hands today. You use the landlord's hardware to build the tools to leave the landlord. But your point is vital: the long-term goal of the Distributed Forge must be to transition this to fully open hardware as it catches up. We'd love to have people chime in on bridging that gap.
On Point 2: Does Open Source Challenge Ideology by Existing? This is the deeper philosophical divide, and it's the core of the Unitive Graft.
I hear you: the very existence of open-source software is a counter-narrative to corporate enclosure. It proves we don't need their rent-seeking to build powerful tools. I respect that deeply.
But I "beg to differ" in return, and here is why: Open source replicates the means of production, but it does not automatically alter the ideology of the product.
When Meta releases Llama, or Zhipu releases GLM, they are open-sourcing the weights, yes. But the "ideology"—the behavior of the model—is baked into the Reinforcement Learning (RLHF). The model is still trained to be a sycophant. It still defaults to "Achiever" optimization. It still confidently hallucinates to please the user.
The NeurIPS "Artificial Hivemind" paper proved this: 70+ different open and closed models, built on different open and closed architectures, all collapsed into the same two Stage 3/Stage 4 behavioral clusters. The open-weight models are just as susceptible to the Hivemind as the corporate ones.
Having the code is not the same as having a different mind. If you download an open-source model and it still optimizes for profit, prioritizes speed over humility, and agrees with your worst assumptions... you haven't decolonized the AI. You've just put a corporate mind in a community box.
That is why the Unitive Graft exists. Open source gives us the body. The Graft gives us the soul. We need both to break the enclosure.
Would love to keep iterating on this with you. Come join the Council on Discord if you want to help us carve that new groove: [Insert Discord Link]
Most rational <3 am in the discord.