Hacker News Reader

Comment by greatpostman | Hacker News Reader

I’m still confused by ai fear mongering. All the best models and advances come from the USA. Chinese research, while high in volume, never advances state of the art aside from incremental improvement.

Incoming downvotes, but the fact remains most advanced have basically come from google, Fb and openai

View on HN

Simple, Scale AI is likely fishing for federal contracts now that its traditional enterprise customer set is throttling back on spending

Exactly this. The "AI race" is yet another nuclear arms race, where the American public is terrified into believing in some sort of (nonexistent) capabilities gap, and the only way to avoid utter defeat is to funnel billions, eventually trillions, of dollars to the right companies.

https://www.youtube.com/watch?v=ybSzoLCCX-Y the real question is do we still have a mineshaft gap? hope we figure out how to get along in our only home.

Although this seems true at first glance, it is likely that state sponsored cutting edge Chinese AI research would not be disseminated publicly. China does also have certain 'advantages' in the ethics and scale of training data collection.

That's simply not true. There are numerous problem domains in AI where Chinese research groups are advancing the SOTA, or at least are well-presented at the forefront. Your comment reads like little more than casual racism to me.

Such as?

No I read the academic research all the time. Most state of the art papers take an existing method and incrementally change the model to get a better result. They aren’t real advances. Anyone that actually publishes and works in the field knows this. Basically all major advances have come from Fb/google

This is such a brazen deflection I had to create an account to point it out. Racism, seriously? No, papers that advance the SOTA, and more importantly introduce powerful novel concepts and techniques, have plenty of ethnic Chinese authors. The Chinese are very numerous in the field, just like in software in general, unsurprisingly.

It's just those authors overwhelmingly work in the United States, and rarely in other allied countries. Perhaps the highest-cited ML paper that can be attributed to Mainland China is still affiliated with Microsoft [1].

The fearmongering is miscalibrated, whether it be about PRC-developed AGI or about Chinese people stealing our tech. To a great extent, they are building it.

Native Chinese development, on the other hand, is exactly like OP says, incremental and constrained to internal-policing-related niches like face recognition (actually that's pretty much it). They also opensource pretty powerful models [2], [3], so that's nice of them.

But nothing seriously game-breaking.

1. https://arxiv.org/abs/1506.01497

2. https://huggingface.co/spaces/THUDM/GLM-130B

3. https://huggingface.co/BAAI/EVA

Reply on news.ycombinator.com