Yes, especially in this atomized society we live in. People have a distance from each other. It’s harder and harder to find genuineness and sincerity. So why not fake it with AI? It’s a sad reality we live in.
In the US, there are communities of people who run small open models locally on laptops/rigs (often Qwen) to get around restrictions. (Maybe with small amounts of fine-tuning or downloading from e.g. Civitai).
Am I interpreting it correctly that there aren’t enough people with sufficient hardware setups to do this in China, or is it much more difficult to download a small open model?
I’d just add a bit of nuance on China side: the constraint is less about raw hardware capability, and more about distribution and friction in access.
In practice, most users face friction in downloading and working with open model ecosystems (e.g. VPN requirements, lack of a unified community layer, and fewer open distribution hubs). On top of that, there’s also more sensitivity around openly sharing or discussing local deployment of uncensored models, which makes the ecosystem less visible and more fragmented.
So technically capable users definitely exist, but it hasn’t really formed into a “default grassroots workflow” in the same way.
Instead, most people still interact with models through apps or APIs rather than local setups.
please love me... is the central meme of human interaction either with AI or other humans
Yes, especially in this atomized society we live in. People have a distance from each other. It’s harder and harder to find genuineness and sincerity. So why not fake it with AI? It’s a sad reality we live in.
There's always a darker landscape when NSFW is involved. People always manage to come up with all kinds of strategies to make it possible.
In the US, there are communities of people who run small open models locally on laptops/rigs (often Qwen) to get around restrictions. (Maybe with small amounts of fine-tuning or downloading from e.g. Civitai).
Am I interpreting it correctly that there aren’t enough people with sufficient hardware setups to do this in China, or is it much more difficult to download a small open model?
You're mostly right on the general direction.
I’d just add a bit of nuance on China side: the constraint is less about raw hardware capability, and more about distribution and friction in access.
In practice, most users face friction in downloading and working with open model ecosystems (e.g. VPN requirements, lack of a unified community layer, and fewer open distribution hubs). On top of that, there’s also more sensitivity around openly sharing or discussing local deployment of uncensored models, which makes the ecosystem less visible and more fragmented.
So technically capable users definitely exist, but it hasn’t really formed into a “default grassroots workflow” in the same way.
Instead, most people still interact with models through apps or APIs rather than local setups.