• This is a political forum that is non-biased/non-partisan and treats every person's position on topics equally. This debate forum is not aligned to any political party. In today's politics, many ideas are split between and even within all the political parties. Often we find ourselves agreeing on one platform but some topics break our mold. We are here to discuss them in a civil political debate. If this is your first visit to our political forums, be sure to check out the RULES. Registering for debate politics is necessary before posting. Register today to participate - it's free!

Trump admin to allow resumption of H20 AI chip sales to China: Nvidia

Bucky

DP Veteran
Joined
Dec 5, 2015
Messages
30,260
Reaction score
7,084
Location
Washington
Gender
Male
Political Leaning
Independent

Looks like Jensen was able to make something happen during his China Trip.
 

Looks like Jensen was able to make something happen during his China Trip.

Why would we allow our A.I. Technology to be given to China?
 
Why would we allow our A.I. Technology to be given to China?

The argument is that the current restrictions may inadvertently push China to accelerate its own domestic chip industry.
 
Thanks.

So instead, we give ours to them?

Makes no sense.

This is the CEO's argument:

He argues that AI technology would be best built on an American technology stack:

China is also already using Nvidia's chips... illegally.

 
Thanks.

So instead, we give ours to them?

Makes no sense.
Far be it from me to disagree with a luminary such as @Bucky, but yes, a critical issue is that these accelerators can be used to train models that accelerate the design of more efficient and performant ICs. Probably the single most promising application of “AI” is toward efficient scalable circuit design.
 

IOW, using A.I. to advance A.I.
 
IOW, using A.I. to advance A.I.
As someone now firmly working in this space, I can guarantee you that there is no wider adoption of generative AI tools than within organizations that are developing AI tools. Heck, I can generate test sets in an hour or two that previously would have taken a large team many months to construct. Humans are still required to QA inputs and outputs at each stage, but human screening is much faster than manual curated generation.
 
Last edited:

You might be just the guy that can answer this question:

"Can A.I. evolve to self-producing, exceeding our capacities to understand or contain it?"
 
You might be just the guy that can answer this question:

"Can A.I. evolve to self-producing, exceeding our capacities to understand or contain it?"
Hmm. This is very complicated to answer.

I would argue that a number of AI models have already evolved to a point where our understanding of how they operate can be limited. Many models embed telemetry to extract insight into their operation - you can see this when, for example, Gemini Pro describes its reasoning. That's useful information for a user, but it's actually essential to the engineer fine-tuning the model.

Example, I just fed Gemini 2.5 Pro the following: Find and solve a problem for me.

Without sharing its actual output, here's the reasoning outputs it produced:

Now imagine if the reasoning itself was not a form of telemetry to shed light on the model's behavior. How would you know -- in the slightest -- how it got from my vague input to the output which was a cannonball trajectory problem? You wouldn't know where to start. So, the Understanding part of your question is already a grey area and we only understand the operation of these larger models to the degree that we internally instrument them.

Containment, I worry less, at least in the near term future, mainly because containment ultimately is a physical act. These models are many things, but they have little physical instantiation in the real world except in the form of motionless racks of compute that are already pretty damn contained. I think there is a scenario where a sufficiently advanced AI that can manipulate a sufficiently sophisticated robot becomes genuinely scary. I'm not too worried about that scenario right now, because I think the AI side of that problem scenario may be only a decade away or less, but I'm certain the robotics side of it is far, far off still. But by the year 2100? Yeah, things could get weird.
 
TACO strikes again

The argument is simple. China becomes dependent on Nvidia, an American company, rather than developing its own Chinese tech stack that could rival or surpass what Nvidia offers.

China will attempt to create its own chips. Other American companies, such as Amazon and Apple, are also developing their own chips to avoid paying Nvidia, but Nvidia has a 3-5 year head start with their superior software. Jensen Huang's argument - Allowing them to sell to China brings in revenue to America and creates more jobs here.
 
Thanks.

So instead, we give ours to them?

Makes no sense.

TACO strikes again

This, again, is a masterclass of the influence and brilliance of one Jensen Huang.

Huang immigrated from Taiwan to America, was sent to Oneida, Kentucky, and was accidentally sent to a military reform school, where he was bullied and beaten up by troubled kids with pocket knives. He eventually became a leader among the troubled kids, where he taught his 17-year-old illiterate bully how to read in exchange to learn how to fight.

The man is a relentless worker - he works every day from 5 am to 9 pm, with no days off, work, work, work. He has now built maybe the most important company in the world.

When Jensen speaks, even Trump listens. The man is a wizard!!!
 
The argument is that the current restrictions may inadvertently push China to accelerate its own domestic chip industry.
Uh huh

I don't have a big issue with NVIDIA selling its lower-end AI chips to China. However, they are already working around those chip limits (e.g. Deepseek), and most certainly will NOT slow down their domestic chip development efforts just because they can buy NVIDIA chips.

On a side note, Thomas Friedman pointed out a few months ago that China is investing heavily on R&D. While Trump is shanking science funding, and chasing STEM students and workers off with a pitchfork, Huawei built a $1.4 billion 2,600 acre research center with labs for 35,000 scientists and staff.



Anyway. Even if this is a good thing overall, this is incredibly hypocritical of the current administration, which has spent years screaming about how evil China is, how we can't let them get their hands on advance technology, and so forth. If Biden had approved this, they'd be shrieking that he was a Chinese puppet. (Not that anyone should be shocked by this administration indulging in such rank hypocrisy.)

I can only assume that Jensen Huang figured out which Trump administration officials to bribe.
 
This is the CEO's argument:

He argues that AI technology would be best built on an American technology stack:

China is also already using Nvidia's chips... illegally.
The CEO of Nvidia thinks it's better if we sell Nvidia chips to more people, news at 11.
 
Cookies are required to use this site. You must accept them to continue using the site. Learn more…