[kevinqi]

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.

[mring33621]

People that think the Chinese are only able to copy western tech are in for a wakeup call.

Actually, that has already happened in many domains, it's just that most western people (USA especially) won't admit it.

[kevinqi]

my assertion isn't that china isn't able to surpass western AI. I think it may well happen. I've been to china many times and am well aware of how ahead they are in many technological/societal areas.

at the same time, I don't buy the idea that distillation is unimportant in assessing what Chinese labs are capable of. If it wasn't, why did Kimi's release timing coincide so well with Fable's launch?

and if Anthropic hadn't released Fable, would we have Kimi today? If the answer is no, then I think that's still a very important point to consider.

[ericmay]

Spot-on and you're asking the right questions. And the other problem in these comments is that folks seem to think if China pulls ahead we can't just distill their models, provided that distillation is a key part of "this". If it's such a great strategy we'll just use it too if we want to. Boom roasted.

For some reason folks seem to think that China can take action and then other countries can't also take action or respond to that action and it comes up again and again. China has hypersonic missiles! Pack it up boys time to go home. Nothing we can do. Dang shucks. China distilled American AI models, welp time to just close it all down and let's just write off those trillions of dollars and all the literal geniuses financing and building these things. Oh well China can just copy American models while we spend all the money! Ok we just stop developing models and we'll just copy their models. China will flood the market with their cheap products! Nope can't do anything like, oh, idk, not buy any of those products or just raise the prices on them in local markets. It's never-ending. I don't understand the lack of capacity to reason about other actors that takes commonly takes place. And that's just China, never mind other general issues.

[Daishiman]

> and if Anthropic hadn't released Fable, would we have Kimi today? If the answer is no, then I think that's still a very important point to consider.

That works both ways, competition and performance spur new developments. You don't think the American labs are looking at Chinese research on how to reduce compute per token?

[nylonstrung]

So many of the breakthroughs and architecture that make LLMs powerful in general today came from China, especially ones related to sparsity and MoE that have made inference and training substantially cheaper.

Let's not forget how much people talked about "prompt engineering" before Deepseek mainstreamed the idea of thinking mode which is now universal

[overgard]

I think it depends on where you think we are on the S curve of intelligence growth. (Yes, I think it's an S curve, not an unbounded exponential). If you think we're near the peak than playing catch up (especially if you can play catch up quickly) is very rational.

I know this isn't exactly a scientific test, but I had a local Qwen 3.6 27B model implement a fairly sizable feature today. There were a couple of bugs, mostly around me not giving sufficient specifications, but they were ironed out quickly when I pointed it out. I was able to ask the model to create instructions so next time it doesn't fall into the same pitfalls, and it did a great job. 27B local model! (And it was super fast too).

I ran Fable 5 as a code review and it didn't really have any significant corrections.

I guess my point here is that, for most work the frontier models are probably overkill anyway, and improving on overkill in a way that raises prices significantly is probably not a winning strategy.

The only place I can think of where the super high powered models are "required" is if you want to do a ridiculous token burn like GasTown where you just have it run un-monitored on very long tasks. To me though, that's an experiment, not a real workflow. And the way these labs are like "oh we made this (broken) thing in a week using just agents!" always also follows with "and it cost $100,000+ in tokens!". Like, ok, I get it if you're doing research but that's the salary of an entire person.. that can actually learn and improve.

[overfeed]

> But the more interesting question is whether labs like Moonshot can be a real competitor to OpenAI/Anthropic.

The answer depends on whether you think the AI researchers at Chinese labs are (or can be) as smart, motivated, and as good at math as those working at US labs - a not-insignificant proportion of whom are Chinese nationals.

[himata4113]

My entire point was that this was not achieved purely from distillation and claiming that is slander against open research.