New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
SOTA models begin their attack on open weight models
https://www.axios.com/2026/07/22/openai-anthropic-open-models-trump-china
They try to make distillation as “ip theft” and illegal. So, it begins 
sota providers see kimi and other models as real existential threats.
Thanked by 1buggedout
Comments
Let's go!
McDonald's lobbies against home cooking burgers, OMG!
SOTA models scrape the entire web, download books and torrents illegally, the entirety of github where the "allow us to train AI with your codebase" is enabled by default, grok build was uploading entire codebases with secrets to their servers just to train their models. But if chinese models distill from SOTA models thats "ip theft"
Honestly I find it very interesting that the expectation from almost everyone is that China is free and has every right to ban US models (and they have been doing so from the beginning), while the US is expected to let Chinese models use American models to create SOTA models cheaper and with absolutely no domestic competition from US labs.
Let's assume that stealing IP is bad.
Then OpenAI doing it is bad.
The fact that companies can go after OpenAI (and have been doing so) and winning settlements is good.
And so Chinese entities creating models doing the same thing, but that you can't go after them for this in the western world, surely by the same definition, that's bad?
It is "bad", I'm just saying that OpenAI/Anthropic should not be the ones complaining, if I copied my homework from John and his friends without their permission, I can't complain if Zhang Wei copies from me.
The problem is, that sota with “think about children” cover tries to hit open weight models to squander posibility to escape their walled garden. Just like that, in broad day light.
Everyone knows that sota steal data, dumping by petabytes. Right now my server suffering from their damn crawlers. This is about blatant attack! And fckin trumper will bend.
I mean, what they're doing is basically moving fast and breaking things. Sillicon Valley motto. They probably expected from the beginning that using all this content will come at a cost, and I assume most AI labs are willing to settle most claims and pay for their training data, at least to some degree. How much they have to pay it probably debatable and it's not like there was president.
Chinese labs have none of that. They expect to pay nothing because China protects them. They have no rights to respect IP. When an author goes to OpenAI and demands money they have a civil case for this. If they have rights to money, they will get money. Either by agreement or as decided in a courtroom by rule of law. The same writer will get absolutely 0 and nothing from Moonshot AI.
OpenAI makes it very easy to block their scrapers: https://developers.openai.com/api/docs/bots
Chinese AI labs on the other hand very likely actively circumvent limits by using western residential proxies, bank cards, company details, etc, in order to actively bypass restrictions while being protected by China doing it. The very same that leverages saving costs this way by then open sourcing these models gain US market share using the resources from US CAPEX.
At the end of the day, it's just politics, China will do everything to win this AI race, such as banning American models domestically, and I think that the US has every right to do what they deem nessesary to accomplish the same, such as by protect their investment by not allowing their enterprises to use these models.
Its not stealing their IP, they're just talking to openai and claude chatbots
What if OpenAI almost word for word can quote copyrighted material, I assume this is what you consider IP theft?
But then Chinese labs do the very same, it's not? Only because they got the data from OpenAI (which they btw didn't admit to)?
The difference being if you got the data from a primary or a secondary source? What if OpenAI didn't get the data from the primary source either?
Lastly, I think Chinese labs obviously scrape copyrighted material, just like American ones do, they're not just training off of American models.
The answer is simple: Community shall stand up for the community.
Users (like myself) who enjoy using open weights models shall contribute to the developers, using the models with "improve=on" so the providers have plenty of real-world data to train and fine-tune models.
This, of course, if the provider is a serious provider and won't abandon the users after they gathered enough data, like Qwen did.
They got massive data input via qwen-code/qwen-cli then, all of the sudden: paywall for code agent.
I get that infra is expensive as fuck, but at least give a free tier to the community that made you who you are, by feeding terabytes of datasets every month.
As I see it only lawyers win in the end
They're getting to taste own poison
and they are hating it.
Jensen Huang Backs 'Excellent' Chinese AI Models, Tells Washington to Ditch 'Science Fiction' Fears: 'Misunderstood The Impact of Kimi'
https://www.benzinga.com/markets/tech/26/07/60642935/jensen-huang-backs-excellent-chinese-ai-models-tells-washington-to-ditch-science-fiction-fears-misunderstood-the-impact-of-kimi