Those who like Senra's talk on Dylan's Chronicles might like my post on Dylan's idiosyncratic reading, "Bob D. on Thucydides." https://substack.com/home/post/p-91303767
Do phones count as a distribution channel? Samsung seems remarkably well-positioned as both a chip and phone maker—if only it could sort out its current troubles.
I think they do, yes. It's probably better to be in Apple or Google's position and own the whole OS, but Samsung has a lot of control about defaults on their phones and that gives them some scale (though not as much as the biggest players -- and enterprise scale may monetize better at first, and eventually consumer scale will probably dwarf it because of sheer size, even if each user monetizes at a lower value.. but if costs can be brought down enough, it won't matter).
Great post! I wanted to offer a slightly different perspective on DeepSeek's open-source strategy, drawing on founder Liang Wenfeng's own words. The common assumption is that openness = vulnerability, but I think DeepSeek is playing a different game.
It's not about easily copied code. Liang makes it clear that the real moat isn't the technical architecture (which, yes, can be replicated), but the team and culture they've built. He says, "the moat formed by closed source is short-lived." Instead, they're focused on accumulating know-how and fostering a bottom-up innovation culture within the organization. Think NVIDIA – their GPU designs aren't secret, but their ecosystem and pace of innovation are incredibly hard to match.
DeepSeek's openness, in this light, isn't about giving away the store; it's about attracting talent and building a community. Liang calls open source a "cultural behavior" that attracts collaborators and enhances their reputation. It's a talent flywheel.
And finally, their focus on AGI isn't about incremental improvements; it's about fundamental breakthroughs. While others might catch up to today's models, DeepSeek's aiming for originality, as Liang puts it. Plus, their architectural innovations (MLA, MoE) give them a massive cost advantage – reportedly 1/70th of GPT-4 Turbo. That's a significant edge.
It's less about secrecy and more about speed and a unique culture of innovation. Liang says, "China must have someone standing at the forefront of technology," and I think DeepSeek's approach is designed to ensure they are that someone.
I highly recommend checking out this interview: https://arc.net/l/quote/rfvuedcq. It really sheds light on their long-term strategy.
The way I look at it, they still made a trade off, but it was a good one for them. They traded off others catching up to their innovations faster for a lot more users, recognition, and goodwill with other AI labs, and become more of a standard globally. That's a good trade even if you don't take into account any higher ideals about openness or open source.
The difference between Nvidia and DeepSeek is that Jensen is ultimately in charge of Nvidia, but if the Chinese government tells DeepSeek to do XYZ, they have to. That's why I wrote bout the danger to their culture of getting all this attention. It's a double-edged sword. So many Chinese founders have been sidelined or reprimanded under Xi that I wouldn't be surprised if this also was a problem here. The government wants a monopoly on power and influence, so anyone who becomes too influential and powerful tends to get cut down in some way.
The AI section... brilliant! 💡
Thank you my friend! Hope you are well
Those who like Senra's talk on Dylan's Chronicles might like my post on Dylan's idiosyncratic reading, "Bob D. on Thucydides." https://substack.com/home/post/p-91303767
Do phones count as a distribution channel? Samsung seems remarkably well-positioned as both a chip and phone maker—if only it could sort out its current troubles.
I think they do, yes. It's probably better to be in Apple or Google's position and own the whole OS, but Samsung has a lot of control about defaults on their phones and that gives them some scale (though not as much as the biggest players -- and enterprise scale may monetize better at first, and eventually consumer scale will probably dwarf it because of sheer size, even if each user monetizes at a lower value.. but if costs can be brought down enough, it won't matter).
Great post! I wanted to offer a slightly different perspective on DeepSeek's open-source strategy, drawing on founder Liang Wenfeng's own words. The common assumption is that openness = vulnerability, but I think DeepSeek is playing a different game.
It's not about easily copied code. Liang makes it clear that the real moat isn't the technical architecture (which, yes, can be replicated), but the team and culture they've built. He says, "the moat formed by closed source is short-lived." Instead, they're focused on accumulating know-how and fostering a bottom-up innovation culture within the organization. Think NVIDIA – their GPU designs aren't secret, but their ecosystem and pace of innovation are incredibly hard to match.
DeepSeek's openness, in this light, isn't about giving away the store; it's about attracting talent and building a community. Liang calls open source a "cultural behavior" that attracts collaborators and enhances their reputation. It's a talent flywheel.
And finally, their focus on AGI isn't about incremental improvements; it's about fundamental breakthroughs. While others might catch up to today's models, DeepSeek's aiming for originality, as Liang puts it. Plus, their architectural innovations (MLA, MoE) give them a massive cost advantage – reportedly 1/70th of GPT-4 Turbo. That's a significant edge.
It's less about secrecy and more about speed and a unique culture of innovation. Liang says, "China must have someone standing at the forefront of technology," and I think DeepSeek's approach is designed to ensure they are that someone.
I highly recommend checking out this interview: https://arc.net/l/quote/rfvuedcq. It really sheds light on their long-term strategy.
Thanks for the comment, great stuff.
The way I look at it, they still made a trade off, but it was a good one for them. They traded off others catching up to their innovations faster for a lot more users, recognition, and goodwill with other AI labs, and become more of a standard globally. That's a good trade even if you don't take into account any higher ideals about openness or open source.
The difference between Nvidia and DeepSeek is that Jensen is ultimately in charge of Nvidia, but if the Chinese government tells DeepSeek to do XYZ, they have to. That's why I wrote bout the danger to their culture of getting all this attention. It's a double-edged sword. So many Chinese founders have been sidelined or reprimanded under Xi that I wouldn't be surprised if this also was a problem here. The government wants a monopoly on power and influence, so anyone who becomes too influential and powerful tends to get cut down in some way.
We'll see! It'll be fascinating to watch 💚 🥃
great stuff. thanks for these detailed thoughts!
Thanks for reading my friend, I appreciate it! 💚 🥃
Us who are about to be obsoleted by Deep research need to go down swinging! haha 😅