[flagged]
All companies are doing all kinds of desperate things to build up their massive training corpus. Perhaps everyone already forgot all the news on the massive copyright violation campaign that US companies have been continuously committing in the name of training models?
> US models are more advanced than Chinese models and don't.
That's only your wishful thinking. As Kimi K3.1 Pro shows, even their open weights models are beating frontier models from US companies. Even if you want to dismiss them for allegedly distilling other models,you would need to explain why they are still beating the distilled model?
I guess I need to test it but I'd put a polymarket: No.
Their old models were basically unusable for agentic AI. Or at least my uses, and I provide AI services to 8 different industries.
At best what you have is a contract violation between the provider and the distiller, and that only applies if the one doing the distillation is the one that entered into the contract with the major LLM provider.
Likely they were smart enough to just not do that. Pay some scrubs on the internet to collect the training data for you. Maybe they violate the TOS, maybe they don't, it doesn't matter to YOU. You didn't have an agreement with OpenAI or Anthropic to violate... And as others have said OpenAI and Anthropic are likely doing the same, which is why ChatGPT sometimes thinks it is Deepseek.