DeepSeek

No single number decides. Layers can disagree — that's the point. New to this? →

• bullish assessment Chamath Palihapitiya

  And part of what made deep seek V3 so successful was that it was open and Western, right. Western companies and Western developers, um, typically don't consume as much Chinese technology from Chinese companies, right? These are kind of separated markets. So by opening the model, China enabled the entire Western world to recognize its capability and build on its technology. And I think it, um, worked so well in its diffusion that then it became a priority for China to have an ecosystem of models.

• bullish assessment Chamath Palihapitiya

  China for a while was considered way behind the United States on intelligence. Right. Uh, and it wasn't until deep seek V3 that it started being treated as a serious contender.

• mixed prediction Unattributed

  At the same time, I could see all companies that are currently open source and closed, right, which includes DeepSeq. I could see them in the future, and Mark Zuckerberg has said he reserves the right not to release all the models in the future, right?

• bullish prediction Unattributed

  And once again, the variants will just go, and everyone will borrow from it.

• neutral prediction Unattributed

  He had two takeaways on DeepSeq. And I was impressed that he landed on both of these. But one was that there was a lot of discussion, especially in Washington, that the U.S. was two years ahead of China. And he said, look, it looks like after the fact that that's hubris, right? If that's now six months, three months, whatever, it's closing.

• bullish fact_claim Unattributed

  And so, yeah, it looks like these guys just decided to be more open. And the MIT license is pretty against the rail.

• bullish assessment Unattributed

  Knowing more about the model allows them to optimize even more. And so, I think the rest of everybody else, other than the big proprietary models, are probably thrilled to have this type of product out there.

• neutral fact_claim Unattributed

  Their compute stack was smaller than the compute stack that OpenAI used to train 01, but it wasn't that much smaller, okay?

• neutral prediction Unattributed

  they can run it well below cost, or they could choose to be running it well below cost in order to have these outcomes. I don't think that's going to last.

• neutral assessment Unattributed

  And they argue about whether OpenAI might just have higher margins or whether DeepSeq might be subsidizing, but that differential's bigger than the one you described for training.

• bullish fact_claim Unattributed

  they believe if you look at the models that are apples to apples on the API right now, that DeepSeq's pricing about 1 20th of OpenAI.

• bullish fact_claim Unattributed

  they talked about the $6 million final training run. And this, it's important to understand, this is also correct, right? And as I said on CNBC, this compares to about $10 or $15 million for O.1 out of OpenAI. So apples to apples, they were 30% to 50% more efficient

• bullish assessment Unattributed

  Number one, he's an incredible founder. Like, there is no doubt about that. He's been working on this problem for upwards of a decade, been thinking about it. He's very successful and has made a lot of money, you know, in the hedge fund business.

• bullish assessment Unattributed

  I think it'd be very hard to make an argument that he's not a remarkable founder.

• bullish assessment Unattributed

  And I think it's impossible to read that and say this isn't an exceptional founder. Like, he's just intelligent, independently minded.

• neutral prediction Unattributed

  But even more importantly, to serve an explosive amount of inference, they would have to spend a lot of money to build that computer.

• bearish criticism Unattributed

  And remind you, DeepSeek spent more than, you know, the amount reported in their last training run.

• bullish assessment Unattributed

  Although DeepSeek surprised everybody.

• bullish assessment Unattributed

  deep seek building great models or they're not going to have rocket companies that copy us and can land themselves like that. That would be naive.

• bullish assessment Unattributed

  we've talked a lot over the last few weeks about DeepSea clearly coming out of left field, very efficiently building a frontier quality open source model.

• bearish assessment Unattributed

  I heard from several inference players that you and I are friends with, that all of a sudden DeepSeek rather than Llama is the enterprise open source model of choice that everybody's experimenting with and playing with.

• neutral fact_claim Unattributed

  Yeah, and only DeepSeek and Grok of all the others have shown the ability to break into the top 10 on the app store download.

• neutral fact_claim Unattributed

  Deep Seat kind of got up there a couple weeks before

• bullish assessment Unattributed

  they're not going to have DeepSeek building great models

• neutral assessment Unattributed

  they still have to compete with DeepSeq on a global basis.

• bearish prediction Unattributed

  they do not want to see a Huawei DeepSeq Belt and Road, either with chips or with open source models. They do not want to see the world run on DeepSeq on, you know, Huawei 910 chips.

• neutral prediction Unattributed

  if that gets traction, you have a window in the U.S. right now where someone might move to go left of DeepSeek in terms of openness on one of their models. Either in an offensive or defensive move. And I think it's a short window.

• bearish prediction Unattributed

  I'm beginning to hear, you know, concerns that D.C. may take action to limit the use of DeepSeek.

• bullish assessment Unattributed

  I believe DeepSeek has been remarkably successful in the enterprise. And that's, you know, it's hosted by AWS. It's hosted by Google. It's being used around the world. I've heard, you know, from Hugging Face that it's been, you know, forked over 1,500 times on their site. And so it's prolific.

• neutral assessment Unattributed

  when we talked about the interview with the DeepSeek founder, I would say he had as much kind of emotional like his his entire emotional mindset was tied to open source, like he believes in it and wants to support it.

• neutral fact_claim Unattributed

  Let's not forget it wasn't that long ago that OpenAI, they'll say borrowed or copied the innovation that happened at Google and DeepSeek.

• bullish prediction Unattributed

  I guess Europe and South America will run on deep seek. Like, because the minute we start vilifying even that ... And if we ban it, boy, it's just kind of, one, you know, even more people working with Huawei chips and deep seek trying to optimize it. But two, you know, in this world we've created where there's so much animosity, I think people will be glad to use it in these other countries.

• bullish assessment Unattributed

  it is, I believe right now, the best performing, most open, you know, if you had a quadrant, the one that's in the upper right on performance and openness

• bullish assessment Unattributed

  I think deep seek has the most kind of intellectual brand because of how that arrived and, and, and almost the, the, the national pride that came along with it.

• bearish criticism David Sacks

  You've got Daario putting out this essay saying we need to pace the frontier days before they launch Claude 5.5 and set a new frontier. I mean, he must know when he's writing this essay that they're about to release a product that will extend the frontier. So, at a minimum, it's hypocritical, but maybe it's even worse than that.

• bearish criticism David Sacks

  I think there's an element of that and I think investors are pretty nervous about giving the founders or or Daria specifically super voting shares.

• bearish criticism David Sacks

  You have current leadership in the company saying that there's a greater than 10% chance of causing human extinction and their own product is unsolved in this area. So think about the risk factors that's creating. How do you IPO with your own management of the company saying that?

• neutral fact_claim Unattributed

  This is a Chinese company actually called deep seek. And, um, they have this new model, uh, very small, only 7 billion parameters.

• neutral fact_claim Unattributed

  but they did move the market and hedge funds love to move the opposite direction because they're a hedge.

• bullish assessment Unattributed

  they did all, all the work to make an awesome AI model and then it happened to be so good that it changed the conversation.

• neutral fact_claim Unattributed

  keep in mind, deep seek is the subsidiary of a hedge fund. That hedge fund subsidiary just moved the market.

• bullish assessment Unattributed

  Maybe they released V3 back in December. R1 came out in January, but it's more like they just had some breakthroughs and pulled off some real innovation.

• bullish prediction Unattributed

  I think it's going to be open source all the way.

• bullish assessment Unattributed

  Great to get a kick in the ass. It's good for the industry. I think America still wins the day. I think open source still wins the day.

• neutral assessment Unattributed

  the rain on everyone's parade from deep seek is still raining.

• neutral fact_claim Unattributed

  it's called, uh, Janus pro seven B it makes images.

• neutral fact_claim Unattributed

  right before we came on air, uh, deep sea dropped a new model.

• bullish assessment Aaron Levy

  And then he, Aaron Levy points out that deep seek is cheap and also blowing up. And he says, all this points to the race is only going to continue to accelerate.

• bearish criticism Josh Kushner

  pro America technologists openly supporting a Chinese model that was trained off of leading us frontier models with chips to likely violate export controls. And according to their own terms of service, take us data back to China.

• bullish endorsement Mark Andreessen

  deep seas work a gift to the world

• neutral fact_claim Unattributed

  So what we have seen here is power of open source. The hosted version of deep seek R one has clear lines around what it can and cannot answer. Clearly president of China off list. Whereas the hosted version does not. What's cool though, is you don't have to use the web one. You can just download it. It's open source and run it yourself and so it strips away all that cruft.

• bullish assessment Chamath Palihapitiya

  software will become more efficient as we saw with r1 using maybe more commoditized hardware less generation of hardware better software better techniques etc

• bearish criticism Chamath Palihapitiya

  this idea that they spent six million dollars probably leaves out hundreds of thousands of illegally imported h100s via singapore that they're not putting into that cost

• bearish criticism David Sacks

  there's substantial evidence that what deep seek did here is they distilled the knowledge out of open ai's models and i don't think open ai is very happy about this and i think one of the things you're going to see over the next few months is our leading ag companies taking steps to try and prevent distillation and so we'll see if uh if the leading ag companies can prevent distillation by third-party companies that would definitely slow down some of these copycat models