Account Info
Log Out
English
Back
Log in to access Online Inquiry
Back to the Top
Nvidia's 2024 AGM highlights: Pay packages and new AI market strategies
Views 5.4M Contents 2331

Does anyone understand $NVDA's sovereign AI pitch?

I thought I did but now think there is some risk this initiative is just capitalizing on current AI enthusiasm, and might result in subscale, duplicative NVDA chips being deployed. Flipside is that I do think that governments will want their own AI infra / models.

The pitch, as I understand it, is that countries should deploy their own AI clouds to 1) train LLM models with texts in local languages and dialects, 2) create research labs / playgrounds so each country can participate in AI innovation, and 3) address sovereign data requirements.

The devil's advocate view of this might be:

1) Aren't all the LLMs already trained on all digital text in all languages? See the screenshots below but seems like there likely isn't much additional digital material any country would have that isn't already in the training corpus. Also, the LLM's already speak different languages.

What content addition would Japan add? Do they have digital texts stashed away someplace that weren't already used? Not sure I see it.

2) Ok, I see the pitch in getting every research lab in every country a GPU factory to see what they can do with it. Could be interesting.

But, are any of these labs that are sub-scale relative to the big US players really going to out-innovate them? Shouldn't they try to compete on post-model applications as opposed to believing they can win in a model war?

3) Every country / region has its own data storage / privacy requirements. But these requirements are already at play in the cloud, and the hyperscalers just make sure the infrastructure they offer is setup to meet all these requirements.

So why do sovereign governments need to do anything more than make sure their data privacy laws continue to be obeyed?

Now, one thing in the bull camp that makes sense to me is government application of GenAI with sensitive data. So militaries may want to train models on their historical data and would want that all to be highly-protected. Likewise, perhaps the IRS and other tax authorities want to train a models on all historical tax returns / audits / etc. to optimize audits going forward, etc.
Does anyone understand $NVDA's sovereign AI pitch?
Does anyone understand $NVDA's sovereign AI pitch?
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only. Read more
6
+0
Translate
Report
17K Views
Comment
Sign in to post a comment
    avatar
    Video Sharer
    708Followers
    68Following
    2115Visitors
    Follow
    More from PREMOSULTRAA