alt Hacker NewsDoes this matter? Distillation is not illegal by every definition of the word.
There are millions of samples available on huggingface and models explicitely trained on output produced by fable. There has been no action taken against them.
Another example is that it appears that the upper limit of what you can do is ultimately dependent on people working on the model, otherwise grok would be a LOT more competitive pre-cursor acquisition.
And lastly, kimi architecture is vastly different than that of fable as it uses mechanisms developed by... kimi themselves. US AI labs are inspired by opensource advancements just as much as open source labs are inspired by traces from models such as fable.
Claiming in any shape or form that fable disillation is one of the primary reasons why kimi k3 is so competitive is slandering the work of other labs that cooperatively push the open-source models forward.
edit: (moved this to bottom) The only argument they have here is that they use GB300 GPU's which for some reason should not be available to chinese citizens.
Replies
Boy I tell you, I am having an awful hard time summoning pity for the organizations that have themselves distilled all of humanity's knowledge into mysterious labor-market-masticating black boxes.
It matters because everyone imagines the inevitable "closing of the gap" between closed and open source, but the rate at which open source catches up with closed source seems to depend on being able to train on and distill the outputs of open source models. As long as performance of open source models is at least partially dependent on frontier-model outputs, then that gap will remain in place by definition.
>Claiming in any shape or form that fable disillation is one of the primary reasons why kimi k3 is so competitive is slandering the work of other labs that cooperatively push the open-source models forward.
If the distillation is irrelevant to why it is competitive, why do they do it then? Obviously is helps improve their benchmarks/performance to some degree, otherwise they wouldn't need to do it.
I agree distillation isn't illegal; I also think Moonshot/Kimi is very impressive. But the more interesting question is whether labs like Moonshot can be a real competitor to OpenAI/Anthropic. If you can only play catchup (however quickly you do that), then you're never going to be at the frontier - I think that's why distillation matters.
> The only argument they have here is that they use GB300 GPU's which for some reason should not be available to chinese citizens
Note that Chinese companies are free to rent from GB300 clouds internationally. There are large datacenter hubs in Singapore and Malaysia serving chinese and other customers.
Though there is also reported [1] significant smuggling of Nvidia chips into China as well.
Legal, illegal…
The word I would use is inevitable. It reminds me of the (PC) clones wars…
It matters because it means that lab could not train that model without distilling another frontier model, and their progress would slow once they get properly cut-off. If I funded that lab, I would want to know that.
Of course it matters. Regardless of whether distillation is legal, there is a difference between training a model with and without distillation. For one thing, the distilled model wouldn't exist without the model it distilled.
Also, companies that use distillation may be competitive but seem unlikely to surpass the companies that are training these models from scratch.
It is incredibly important to whether the US can maintain its AI lead. If foreign competition is closing the gap only by distillation, then the frontier labs can focus on preventing distillation and maintain their lead that way.
US dominance is also important for approaches to safety, especially political approaches. If the frontier models are all US-based, safety might be tackled via internal US policy. If other countries can independently train competitive models, international cooperation is required.
Edit: It is also important for the business model. Companies won't be able to justify tremendous training costs if competitors can replicate their product much more cheaply via distillation.
It certainly matters as familiar sounding words to their stock holders to please not drop the valuation.
Because what they want them to think is "the AI factory has unique proprietary technology that cannot be replicated"
What they don't want them to think is "it's relatively easy once you know the basics to bootstrap to near SOTA and so the commercial case for selling inference has an extremely short profitability horizon with little if any brand loyalty or lock in".
I wouldn't be surprised at all if US labs are also distilling Chinese models, except we'd never know since they can simply self-host them
It does matter in that these LLM companies need to be run into the ground, and every embarrassing clod working for them run out of town.
It's showing that 'distillation' is a viable way to reclaim all of what they stole and hoard, and with enough luck their debts will come due in time for them to feel it.
> Does this matter? Distillation is not illegal by every definition of the word.
Correct, but it at least helps answer the question of "how do they make such good models for a fraction of the price???" The answer is someone else spends the untold billions and Chinese labs do a little tweaking.
It doesn't matter. It is most likely a pretext for upcoming actions mostly likely executed via yet another retarded executive order. The guy that posted this looks like he's drowned himself in the MAGA Koolaid.
It also doesn't matter for a simpler, and much grander reason.
All LLMs are trained on the corpus of humanity's knowledge, the legacy of everyone who's ever lived and our civilization as a whole.
Anything that prevents or circumvents the accumulation or gatekeeping of this knowledge and puts it in the hands of more people (that are not AI company shareholders) is a good thing. Whether that is done by open sourcing the model weights, the training set, or by making the output better and cheaper, it is all fair game and is, as another poster mentioned, inevitable in the long run.
[dead]
Uh, what?
> Distillation is not illegal by every definition of the word
Note that Anthropic (and USG) alleges [0] not only that Kimi was distilled, but that they actively circumvented measures intended to stop distillation. There are multiple ways that's illegal, including:
- Civil breach of contract. Anthropic's TOS explicitly say you can't do what Kimi is alleged to have done.
- Economic espionage: 18 U.S.C. §1831 criminalizes obtaining a trade secret through theft, fraud, or deception while intending that it will benefit a foreign entity.
- Trade-secret misappropriation: if Anthropic could argue industrial-scale querying reconstructed proprietary aspects of Fable (like by showing it produces similar outputs, as others have done) then it's illegal under 18 U.S.C. §1832.
- California computer-access statute §502 bars knowingly accessing a computer system and, without permission, taking, copying, or using its data.
- Computer Fraud and Abuse Act protects against the case where restrictions against an activity are circumvented (like Kimi is alleged to have done).
> There are millions of samples available on huggingface and models explicitely trained on output produced by fable. There has been no action taken against them.
A lack of prosecution does not make something legal. There is also the scale/commercialization thing, which isn't an issue with random tiny HF datasets/models. Remember: Kimi also sells K3 inference.
> kimi architecture is vastly different than that of fable
How do you know that? Do you work for Anthropic? Also, this has nothing to do with architecture, we are talking about data.
> US AI labs are inspired by opensource advancements just as much as open source labs are inspired by traces from models such as fable.
Cool. The difference is that one of those things is legal (because they chose to open-source) and one of those things is illegal theft of trade secrets (because it was stolen).
> Claiming in any shape or form that fable disillation is one of the primary reasons why kimi k3 is so competitive is slandering the work of other labs that cooperatively push the open-source models forward.
1) this has nothing to do with other labs, just Moonshot (and Z.ai, MiniMax, DS)
2) slandering or not it happens to be completely true, so, there's that
[0] https://www.anthropic.com/news/detecting-and-preventing-dist...
It matters because the closed-source frontier labs spend lots of money on human data (RLHF / RLAIF with human oversight). Moonshot is accused of circumventing these costs. Frontier labs add research costs into their inference pricing. If the market doesn't permit them to sustain sufficient pricing to have a positive cash flow, then their business prospects become weaker and they risk insolvency. Furthermore, other leveraged companies are at risk.
The reason why the United States government is weighing in is because it's in the national interest of the US to have supremacy in "AI".
Legality or lack thereof is one of many data points about whether a thing is noteworthy.
Moonshot performing distillation is rational from their point of view. Reducing costs is in the interest of businesses. It's also rational for frontier labs and the US government to add obstacles to this process.
As consumers this is probably a positive development.
Perhaps even more importantly, the current frontier LLM models are self-admittedly the product of enormous quantities of copyright infringement and even less savory inputs, so calling them out for distilling the fruit of that tainted tree reads as highly hypocritical at best.