Live data from Hacker News

On DeepSeek and export controls

darioamodei.com

61–70 of 185 posts

Re: On DeepSeek and export controls

#61
post #10

Unless the author makes a compelling case about why AI breaks the MAD status-quo between nuclear powers, I will assume that their appeals to NATSEC are an attempt to artificially create moat for their company.

I also find it counterintuitively reassuring that we have already had the power to blow up the planet for decades. But at least in theory - GPT10 level AI could help develop missile defense systems that actually work at scale, which would eliminate the MAD status quo, no?

Re: On DeepSeek and export controls

#62
post #44

Earlier quoted context omitted.

Absolutely. Chinese companies shouldn't have chips because their government has "committed human rights violations, has behaved aggressively on the world stage". And the US government hasn't?

Just wait. By year 2 of Trump 2.0, China is going to be the global good guy that Europe and Latin America make their new best friend.

Year 2? We're almost there at _month_ 2.

Re: On DeepSeek and export controls

#64

I'm kind of tired of the entire narrative that "US good, everyone else bad", as if only US deserves to hold powerful technology because it will be used for "the greater good", rather than employ it in military applications, as if Anthropic didn't partner with Palantir. And it amazes me how many people can't seem to see past the "US is good, everyone else are bad" smoke mirror.

I live here, so I want it to be the best.

I'm not sure how that relates to original comment. Do you mean you want everything that is or could be better than American technology banned/destroyed so we stay the best...?

Like, any global hegemony will be increasingly corrupt given the power that gives, IMO.

Re: On DeepSeek and export controls

#65

I'm kind of tired of the entire narrative that "US good, everyone else bad", as if only US deserves to hold powerful technology because it will be used for "the greater good", rather than employ it in military applications, as if Anthropic didn't partner with Palantir. And it amazes me how many people can't seem to see past the "US is good, everyone else are bad" smoke mirror.

First- the author of this blog post is the CEO of Anthropic, an American AI company. Of course they are going to argue for export controls if it can hurt their competition. They get to benefit from anticompetitive practices with none of the legal risk! So it's not even about keeping the "bad guys" from having things, it's really about making more profit. Also, you can typically assume that anything a CEO publishes is in a pursuit to raise stock prices, it has nothing to do with morality. If CEOs were moral people most of them wouldn't have become CEOs (there are probably a few CEOs that are truly good people, but I'll bet there are more "wolf in sheep's clothing" CEOs that use an image of morality to improve their company's reputation).

Second- aside from the specifics of this post, you don't have to believe that the US is good to see the logic of export controls. There are a number of countries that are openly hostile to the US (Iran and North Korea), and also some that are semi-privately working against the US (like China). These countries will take any advantage that they can- they will steal IP and use the US's development resources to catch up to the current state of the art at a reduced cost. So you don't have to put this in terms of good vs. bad, just think of it in terms of things that benefit the US and things that don't benefit the US. Whether you think people are evil is irrelevant, it is very logical for the US to put export controls on things that it doesn't want to give away for free. The US wants to preserve its advantage and make it as costly and difficult for these other countries to catch up.

Re: On DeepSeek and export controls

#66
Without taking a position on unipolar vs. multi-polar:

Dario makes an astounding implicit assumption:

- China originating labs cannot acquire chips providing 80-90% similar utility without the US within the next 2-3 years.

I'll make an observation, re: DeepSeek's incentives that drove them to create the innovations from the V2 and V3 papers.

DeepSeek, compared to American AI labs, are much more compute constrained, but in a unique way. Their chips are more memory bandwidth constrained (depending on type anywhere from 50% to 80% less bandwidth).

Therefore, each dollar/hour of investment towards memory optimization is worth MORE to DeepSeek than to American labs.

In the V2/3 paper, they've demonstrated exactly that with these memory optimization techniques.

1. MLA -> reduces KV cache by nearly 80% compared to GQA. By the way, this was published in V2 in May 2024.

2. FP8 matmul (while still accumlating in FP32 gradients) without losing significant quality.

3. DualPipe scheduling and reworking of Hopper SM's allocation on communication vs. computation -> DeepSeek's V3 paper has 2 full pages of hardware suggestions for "hardware designers" (read NVIDIA)

Export controls in a global market create different incentives in parties. The resulting incentives will change, and agents (using it as an traditional economics term) will change their capital allocation strategy.

Re: On DeepSeek and export controls

#67
post #59

Earlier quoted context omitted.

I mean, in the end, USA is still better than the alternative.

I don't know. Living in a country where being able to shoot people is an absolute, immutable, inalienable right granted by God himself (more so than anything else in the Bill of Rights) but where access to food, shelter, healthcare and education are all expensive privileges subject to the whims of the free market, and where half of the population doesn't even believe viruses are real, and where you don't even need th…

It does if you're quickly converging into the Handmaid's Tale universe. May the odds be ever in their favor - I'm quite happy we don't carry guns to make it through the day...

Re: On DeepSeek and export controls

#68

I'm kind of tired of the entire narrative that "US good, everyone else bad", as if only US deserves to hold powerful technology because it will be used for "the greater good", rather than employ it in military applications, as if Anthropic didn't partner with Palantir. And it amazes me how many people can't seem to see past the "US is good, everyone else are bad" smoke mirror.

He's unlikely to be writing from a genuinely ideological point of view. He's playing a zero-sum game, in that every chip that goes to China is one that Anthropic and/or its cloud provider doesn't get its hands on.

So it makes sense to him to argue for export controls using whatever rhetorical flexes and flourishes he can come up with.

Re: On DeepSeek and export controls

#69
post #62

Earlier quoted context omitted.

Just wait. By year 2 of Trump 2.0, China is going to be the global good guy that Europe and Latin America make their new best friend.

Year 2? We're almost there at _month_ 2.

Oh, people are thinking it now for sure, but it takes a while to retool your relationships and build out a new trade network so you can safely tell a major trading partner to fuck themselves to threats of 100% tariffs.

Re: On DeepSeek and export controls

#70
post #11

Earlier quoted context omitted.

This is a choice between the lesser of two evils, or what could be called "drinking poison to quench thirst." Export controls might lead China to develop its own advanced chips, but without them, China will certainly gain powerful AI capabilities

Point being, they will gain those capabilities either way. The only question is whether we in the US choose to hamstring ourselves with bureaucracy while we watch them do it.

I think it is even worse than this.

The only way this would work is to create a kind of great firewall of America keeping open source Chinese models out while the remainder of the globe uses something like DeepSeek.

It would literally be the dumbest soft-power move in the history of the US.

Post reply on HN