Hacker News Reader

Comment by Aboutplants | Hacker News Reader

Not worried about that, you will only have to wait 3-6 months and get a Chinese model just as good.

That’s misunderstanding why these models are behind. A large part of why they’re behind is they aren’t able to do the reinforcement learning post-training steps that takes a pre-trained model and turns it into a frontier model like GPT 5 or Opus. Instead they do their best to recreate these models using distillation.

Fundamentally, you can never distill your way to being the teacher, so these approaches will not advance the frontier.

[edit, after thinking about it I think my phrasing is unfair. It's not necessarily that aren't able to do it, but they haven't yet shown that they are willing to do it.]

That’s not remotely true. They did distillation as a cheap solution to the cold start problem. You need data/trajectories to hill climb to higher capabilities. All large Chinese labs do RLAIF.

Oh yes, not remotely true. Which is why the frontier labs all have invested heavily in trying to identify and thwart distillers, using known company names / domains to drive their exclusion lists.

It's cheaper to distill than to do reinforcement learning, so of course they prefer that, but if it wasn't an option they could just pay up and spend more GPU time on RL.

>"they aren’t able to do the reinforcement learning post-training steps"

Not yet.

If there is a need someone will come and fulfill. Personally for me now I do not even want to use top models. Professionally I use AI to help with the coding using Junie agent that comes with IDEs from JetBrains. Junie is told to use Gemini Flash and works fine for what I ("I" being an emphasis here) ask it to do. I tried more advanced models and different vendors only to discover credits going down the toilet without any extra benefit.

I'll agree I guess and clarify that the better phrasing is probably something like "haven't yet shown the capability to."

> you can never distill your way to being the teacher

Are you sure?

What if you distill from 10 teachers?

In this case all teachers have also learned from each other.

I think GLM 5.2 having a higher ELO than Opus 4.8 shows that they did it [https://gptbased.com]

Chinese companies giving away expensive models for free is a symptom of the AI bubble, too. It's not a law of nature that they'll always be able to scrounge up the money for yet another training run.

Shaping the tool that does the thinking is quite valuable when you're in the business of changing how people think - I think we can expect propaganda agencies to be subsidizing model creation forever.

This doesn't strike me as a symptom of a bubble - except in so far as the bubble pushes the competitors models forwards and thus they need to invest more to stay competitive.

All the models, have to respect their local laws, and most of all, pressure from users and the employees.

They all carry political weights, because humans behind defend their interests, and are promoting some social values.

https://pastebin.com/hjhvsBFg

This answer from Claude is so biased that it is ridiculous

I think it's a deliberate business strategy of commoditization of their complement.

China acts like an entire bloc, not as single companies, and they want to monetize hardware.

If you think Chinese companies always act as a bloc, your mental model needs to get about a billion times more detailed. But in this case just a few details may be enough: There are Chinese AI companies that have released LLMs without publishing the weights.

ByteDance is going the direct-to-consumer route with their Doubao chatbot (the most popular in China, probably thanks to their social media prowess). iFlyTek seems to be angling for enterprise and government use cases, where they already have an in.

The companies that have released weights have in common that they didn't have a monetization channel lined up and their models weren't good enough to make people pay attention with just API access. (You can see with Qwen Max that the calculus can change towards not releasing weights for better models.)

And who exactly among the investors is having their complement commoditized? When Nvidia releases Nemotron, the story is clear, but it's less obvious for say Z.ai's GLM.

I have never said they always act as a bloc, but their industry has a strong component of long-term strategic government planning behind them.

> China acts like an entire bloc

> I have never said they always act as a bloc

Pick one

As long as it is in the CCP's national interest to have a frontier model, Chinese companies will have the resources for another training run.

Reply on news.ycombinator.com