Broadcom Inc.

Gavin Baker support ×2 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E238 · 2025-08-09 ▶ 1:30:10

Yeah, it's not comparable to Colossus. So it's just, it's not going to be decisive in this, you know, AI, you know, race that certainly China believes we're in.

Gavin Baker support ×2 ▲ on 🤖 Pre-training scaling laws still hold — Colossus and Grok 3 are the proof E238 · 2025-08-09 ▶ 11:43

But this is the first time they have not released a model that was a decisively better. And the progress for Grok 4 was exceptional. Grok 5 is coming soon, you know, they have more.

Sacks oppose ×2 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E237 · 2025-08-01 ▶ 53:05

these open-source models are catching up pretty fast. We're ahead in many other aspects. Our chips are a lot better. Our data centers are better and so on.

Gavin Baker reversal ×3 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E236 · 2025-07-19 ▶ 1:01:03

And so I think it's a smart decision for America to sell this chip that gives America the advantage in AI and keeps China from developing kind of their own domestic alternatives, which could eventually kind of challenge Nvidia, AMD, you know, other kind of American AI accelerator champions globally. We don't want that to happen. So I thought it was a mistake when they banned the chips. And I'm really happy and think it's really good for America if that is being reversed.

Gavin Baker support ×3 ▲ on 🤖 Pre-training scaling laws still hold — Colossus and Grok 3 are the proof E236 · 2025-07-19 ▶ 20:18

So this is incredible progress. And I think it's worth mentioning, this model was trained on Hopper. So this model is probably about as far as you can take the last generation of Nvidia GPUs. The next models we see, you know, Grok 5, you know, O5 from OpenAI, whatever they're going to call the next Gemini. They will be trained on Blackwell. And I think that will be a really big step function.

That's the global lesson, by the way. Chamath, you're totally right. Conceptual, the blog post is right. But that's only true when you have enough data and depending upon the use case, the level of data you need may not be possible for years, decades, and you may need to hack your way there through human interactions.

If you get, if you have an LM or foundational model of some kind that is the best in the world of the scientific method, game the F over. You basically, you just light up more GPUs and you just got like a thousand more PhD students working for you.

Chamath support ×3 ▲ on 🤖 Pre-training scaling laws still hold — Colossus and Grok 3 are the proof E235 · 2025-07-11 ▶ 20:39

So why is this crazy? Well, he made this huge bet on this 100,000 GPU cluster. People thought, wow, that's a lot. Is it going to bear fruit? Then he said, no, actually, I'm scaling it up to 250,000. Then he said it's going to scale up to a million. And what these results show is a general computational approach that doesn't require as much human labeling, can actually get to the answer and better answers faster.

Friedberg oppose ×3 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E232 · 2025-06-21 ▶ 37:37

So I do think that the lithography IP moat is being crossed in China. I do think that China is developing actually new technology for DUV and EUV systems. I do think that there's a risk to Nvidia's core. ... we're creating every incentive for an alternative to Nvidia to emerge from China.

I think Sacks' work on the diffusion rule, just generally, I don't think has gone enough attention in the rescinding of the diffusion rule, which essentially handicapped our ability to even arm our allies with our semiconductor technology.

Sacks support ×3 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E229 · 2025-05-24 ▶ 1:04:45

I think that most people in Washington believe that the export controls on China make sense, because we simply can't allow our most powerful semiconductors to be used by China because of the dual use. Now, where I agree with Jensen is that I think we should not place undue restrictions on the rest of the world using the American tech stack, because we will simply concede the market share to China.

I think we should keep them dependent. Keep selling it to them.

Friedberg oppose ×2 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E224 · 2025-04-19 ▶ 11:20

So, Jay Cal, I think they're doing it either way. I think that they're going to invest and build their own EUV and chip manufacturing capacity either way.

Chamath support ×3 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E224 · 2025-04-19 ▶ 14:20

I think that the technology that they need is extremely non-trivial, and I do think that it actually slows them down quite a bit if they don't have access to it.

Sacks support ×3 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E224 · 2025-04-19 ▶ 7:57

It seems to me that at some point you have to say that some technologies are just too sensitive to be sold to China. And so then the question is, are you drawing the line in the right place?

Friedberg oppose ×3 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E222 · 2025-04-05 ▶ 1:18:44

The components to do the manufacturing, so there's a company called ASML, which everyone that works in tech knows about. They have these advanced lithography systems to make the semiconductors. They don't sell to China, so now China's got their own homegrown lithography systems.

Chamath oppose ×1 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E221 · 2025-03-29 ▶ 18:46

I think they're doing it now. They're trying now.

Gavin Baker support ×2 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E221 · 2025-03-29 ▶ 18:57

In the next five years, I think zero. I think it's really, really hard. But over the 10 years, who knows? And if you're the CCP, 10 years isn't that long. If you're America, 10 years is an eternity.

while China is, you know, their kids are the leaders in AI. You know, DeepSeek should show us that, you know, we have underestimated China every step of the way.

Chamath oppose ×1 ▲ on 🤖 Pre-training scaling laws still hold — Colossus and Grok 3 are the proof E218 · 2025-03-08 ▶ 6:45

Open AI dropped GPT-45, and I think it was not really that well received.

Jason oppose ×2 ▲ on 🤖 Pre-training scaling laws still hold — Colossus and Grok 3 are the proof E218 · 2025-03-08 ▶ 1:24:30

Let's talk about this ChatGPT. I guess they came out with 4.5. It was such a dud. I didn't even realize that they launched it.

Jason support ×1 ▲ on 🤖 Pre-training scaling laws still hold — Colossus and Grok 3 are the proof E216 · 2025-02-21 ▶ 1:29:36

It's super consequential. I mean, complete agreement with Chamath

Chamath support ×3 ▲ on 🤖 Pre-training scaling laws still hold — Colossus and Grok 3 are the proof E216 · 2025-02-21 ▶ 1:19:36

I mean, basically what he proved was that there are still valuable gains in pre-training, and so the larger the cluster, the more value that there is... My first takeaway was I was sneakily surprised on the pre-training upside on having a larger cluster. So I think that that's very pro-NVIDIA actually, and it's actually also just really good in general for foundational model makers.

Sacks oppose ×2 ▲ on 🏛️ US compute export controls keep China off the AI frontier from 2025 E215 · 2025-02-15 ▶ 45:00

I think that the whole Paris AI Summit took place against the backdrop of this recognition, because just a few weeks ago, we had deep sea. And it's really clear that China is not a year behind us. They're hot on our heels or only maybe months behind us. And so if we hobble ourselves with unnecessary regulations, if we make it more difficult for our AI companies to compete, that doesn't mean that China is going to follow suit and copy us. They're going to take advantage of that fact and they're going to win.

The point I would agree with is that it's going to happen anyway. And that's what DeepSeek proved. You can turn off the flow of chips to them, and you can turn off the flow of talent. What do they do? They just get more efficient, and they exported it back to us. They sent us back the best open source model when our guys were staying closed source for safety reasons.