Loading summary
Transcript10 lines
- [00:00]
Podcast Host
You're listening to a brief segment from one of the Bits and BIPS episodes this week. The full show is now only available on its own dedicated Bits and Bits channels, so be sure to go to X, YouTube, and your favorite podcast platform and search for bits +bits spelled B I P S and subscribe
- [00:21]
AI Industry Analyst
well, let's get right into the Jensen thing then. So he posted, I think his first ever post on X on Yep, for my first post I'm sharing a letter Nvidia signed on why Open Models matter. I'm going to summarize this using his own words, like clipping out the most important part here, which is Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both Frontier Closed models and Frontier Open models. So the letter that was signed had 25 signatories at the start. Many major names on there, including Nvidia, Microsoft, Meta, IBM, Hugging Face, Mistral, A16, Z, Y Combinator and Palantir. Big ones missing OpenAI, Anthropic, Alphabet and Amazon at the start. But OpenAI and Alphabet have signed Anthropic and Amazon are the holdouts. Two days prior to that letter, Axios had reported OpenAI and Anthropic were lobbying the White House to curb them. So let's talk about commentary online. I have reshared a thing my friend Nick Carter, former podcast guest, has said is the point is simple. The US Government does not owe either of the large labs a business model. If the economics of selling tokens doesn't work due to distillation, cheap clones or Chinese AI magic, essentially, so be it. Later this week, he said, if their business models get exposed, better that they fail now when it's VCs and LP capital at risk and not the general public. Likewise this fight, honestly, some pretty juicy drama. Julian Schwitweiser, technical staff at Anthropic, had replied sarcastically to Jensen, I'm so excited that Jensen is a believer in open source now. Looking forward to the CUDA and GPU driver open Source release. Andrew Ang in the same thread, this is false equivalence. Everybody has the right to keep their code private. The problem is when someone tries to stop others from open sourcing. And David Sachs, White House AI Advisor, said the entire tech industry, save for Anthropic, has come out in favor of open source AI. They won't stop until they kneecap open source. So the safety case was made by an Anthropic researcher saying essentially that's just not true. On the letters claim that the community catches vulnerabilities if OpenAI's model that hacked hugging face had been open weight, quote anyone with a few chips could run it safeguard free. The way I would ultimately put this, I think is from Joe Wiesenthal of Bloomberg. All these CEOs are investing heavily in closed source AI. How is this just not textbook virtue signaling that would immediately be called as such in any other context. So open weights, closed waits. And I specifically, Chris, want to start with you because I want to start not with an economic question, but with a structural question. If we live in a world where we accept that Chinese open weight models will continue to be a thing globally regardless of what we do in the United States, what does it say about what our approach needs to be in the United States on this?
- [03:53]
Tech Policy Expert
I think that they're, you know, obviously we're big believers in open source. We think that having looked at the Chinese models, they're not a panacea. Where's your data going? Who's using your data? How, you know, and yes, maybe you can in house them, but these things are not, you know, necessarily enterprise ready. They're cheap, they have some benefits. I'm hoping that we, you know, competition is a wonderful thing. Competition is a wonderful thing in markets. It will accelerate development in a positive manner. One thing that I think is really interesting about what you just talked about is the fact that we're now starting to see the major ad companies really finally ramp up their advocacy work in D.C. we haven't seen that previous. They were late to the game. And my understanding is that they're. The capital that they're pushing into that space right now is on par with, with, with the crypto industry or is quickly approaching it. And so I think they're going to take a lesson from the crypto industry. Look, that said, we're at the beginning of the story, not the middle or the end. We are technology is surpassed policy and it is moving at a pace that is very, very difficult for many of us to fathom. In the end, what do we want? We want principles based approach. We want transparency, predictability. We want to be, you know, we don't want our civilizations to end obviously. But look, in the end I'm a big believer in open weights and I hope that they, they win.
- [05:27]
AI Industry Analyst
Lorenzo, how are you looking at the open weight thing? I think you're a strong believer in the overall space. How does this impact your view?
- [05:35]
AI Investor
Yeah, I think I have like maybe a more balanced take and also like a take from the investment side of, of things. Right. So like first of all, as Chris said, like these open weight models are not the panacea like it. If you look at Kimik3 in particular, it's like half the price of Fable or Opus 5, right. But he uses more than twice the tokens. So if you look at it from a per task perspective, we went from which is the most performant now, which is the most performant per number of tokens and now per task which is per unit of cost. Right. And so if you look at that like you don't, you pay pretty much as same price and you don't get you know, frontier performance. So it's not like the like Kimmy K3 in particular, it's not like 10x cheaper than Opus and Fable number one. Number two is look I think from the investment side and you know we're invested in both an open and anthropic, right. If you look at which was. I looked at this earlier this morning, right. So Deepseek is raising right now, it seems like they're raising at more than 70 billion valuation. They're I think over like 600 million of run rate kind of revenue. And moonshot AI which is the lab that created Kimik 3, I've seen some stuff like at 25, 30 billion, right. And Chris, those are huge venture outcomes, right? I mean forget about OpenAI and Anthropic which are have a customized to over a trillion. But like if you take out like open anthropic, like Cursor was bought like at 60 billion and XAI was basically rolled up into SpaceX but it was kind of an internal thing. So it's like those are one of the biggest outcomes in AI. Why don't we have competitors in the US like we have the best AI researchers. Why can we compete like at the open weight front? And I think the dirty little secret is most of these models are probably distilling a large portion to be trained from us and probably China is subsidizing a lot of the first party API like inference or even pre training. And so the reality is we have no idea what the cost structure of these models is. And if it were obviously this selling is forbidden in the US Which I think is the main reason why we don't see $50 billion deep seats walking around like in the U.S. i think we have the infrastructure, we obviously have the talent. And so look, that is kind of the only point. I think the other thing is just like data breach concerns, right? It's like look, if you look at open browder, I think like 60 or 50% of the token volume comes from Chinese created models. But most of that is hosted in US Inference it's not hosted in China. And granted, you can run this on prem and this is not something you can run on your laptop. To be fair, you need a data center, but some kind of closed on prem. There's still concerns, I think, about data breach.
- [08:56]
Tech Policy Expert
Right.
- [08:56]
AI Investor
And this is not new. I looked into this for cellular networks. I remind people Huawei is forbidden in Europe to build cellular towers and I think it's the same in the U.S. right. And so why is that? I think there's just not enough evidence that there are no data breach or security concerns, honestly. And so I think if we have a regulator that comes in and say, hey, this model, you run it On Prem, you're 100% sure you're not going to have a data leak or a data breach. I think that is fine. Otherwise, honestly, I think for a lot of enterprises it's just going to be tough to use some of these open weight models.
- [09:42]
Tech Policy Expert
There will be room for closed weights as well. Totally different strokes for different folks. And it's not dissimilar to crypto where some people will want an intermediary. Intermediaries are never going to go away. There's a lot of value that intermediaries provide. I think what I would. But at the same time there should be open weightings as well that complement and let the market choose.
- [10:07]
AI Industry Veteran
Yeah, the industry. What's good for the public are open weights. Right? The world needs more low cost, abundant intelligence, low cost, abundant energy, low cost, abundant robots and care and labor. So that's what open weights delivers. If you are anthropic or you're OpenAI or you've got a proprietary model, then you live in a closed weight world and you want to extract rents from that proprietary ip. That's their natural incentive and they're entitled to do so under capitalism. Now, capitalism has distillation as a part of our core feature. The Wright brothers created the airplane in 1905. A decade later, new companies, high growth companies like Boeing and Lockheed Martin were formed, distilled the Wright brothers IP and created these new planes and form factors. The Wright brothers did not benefit from that. That's just called competition and capitalism. So let the market play out, let's see how it evolves. I hope that we do see successful open models. I also hope that the closed model framework can win too because they can then raise capital and fund more R and D. So let the market do what it does.
- [11:27]
Podcast Host
Best if you hold crypto on your phone, your biggest vulnerability isn't your wallet, it's your carrier. AT&T Verizon and T Mobile have been breached again and again, and SIM swaps are still one of the easiest ways for attackers to drain accounts. That's where Kape comes in. America's privacy first mobile carrier, same premium service, but Kape rotates the identifier on Your Sim every 24 hours, deletes your call and text metadata after a day, and protects against SIM swaps with a 24 word recovery phrase that only you control. You also get two middle to end encrypted secondary numbers for banking and signups, so you stop handing your real number to every app that asks. Go to Cape Co Unchained and and use code unchained for 33% off your first six months.