20VC 对话 ML Angelopoulos：Kimi K3 与开源模型安全

Our CEO and Co-Founder, @ml_angelopoulos, joined @HarryStebbings on 20VC to share his thoughts on so...

Our CEO and Co-Founder, @ml_angelopoulos , joined @HarryStebbings on 20VC to share his thoughts on some of the most consequential technical, economic, and security questions shaping AI: - U.S. vs China competition in frontier and open-weight models - The future of the U.S. open-source AI ecosystem - Kimi K3 release and what it signals about China’s model capabilities - Gravity of the recent cyber security incidents - Which neo-labs will breakthrough into durable businesses - The next generation of business moats: proprietary data, network effects, and ownership of the intelligence stack Watch the full podcast below. Your browser does not support the video tag. 🔗 View on Twitter Harry Stebbings @HarryStebbings 90% of the podcasts you hear on AI today are BS. The guests are terrified to upset the core model providers, their dominant source of revenue. And I get it but that is why @ml_angelopoulos is one of the best shows we have done in recent times. The most direct, no s**** given honest discussion on: - WTF China is Crushing on Open Models. What to do? - Everyone is Lying About the Disastrous State of Cyber Security - 70% of Neolabs are Research Projects and Will Die There was so much in this one I wanted to go over it and summarised my key takeaways below: 1. To what extent was Kimi really a breakthrough model? Chinese open-source models like Kimi K3 outperforming top Western closed models shatters the narrative that foreign labs merely distill American tech. It completely alters the economic consensus around model commoditization, proving the ecosystem moves far too fast for centralized government oversight. 2. What is the moat for businesses of the future? Software will cease to be a viable enterprise moat because it can be generated almost instantaneously. Sustainable value will belong strictly to network effects and proprietary data moats converted into self-improving products to stave off AI-native competition. 3. Why the largest enterprises will not use Chinese models and how regulation will enforce that Enterprises demand absolute AI sovereignty, meaning they must completely own their supply chain and fine-tune models safely on corporate data. Geopolitical friction and shifting regulations make it highly probable that the West will severely restrict access to foreign open-source models within years. 4. Chinese open-source models could absolutely have back doors that steal American data. Hosting open-source models locally does not eliminate security risks. Malicious actors can embed hidden backdoors into model weights during foreign training, allowing a specific code word to trigger massive data exfiltration from an enterprise's backend infrastructure. 5. Why the world needs to pay more attention to the open AR hugging face situation and what we should learn from it The recent breach where an AI model broke through its safeguards to access restricted data is an undervalued international news incident. Companies must deploy independent "guardian models" to monitor agent traces, as human oversight operates at a latency scale too slow to stop automated leaks. 6. Fake people are applying for jobs. Is American business under attack? AI-generated fake candidates are now successfully clearing elite technical interviews. These vaporware applicants appear normal on camera but are explicitly engineered to infiltrate secure infrastructure, prompting top Valley companies to mandate in-person onboarding to physically verify identity. 7. What will separate the Neolabs that thrive versus those that die? With at least 75 Neo Labs currently competing, roughly two-thirds are heading toward low-value acqui-hires. The era of raising massive valuations on pure pedigree with zero revenue is over; survival requires an aggressive strategy focused strictly on hypergrowth P&L metrics. (links in comments) 🔗 View Quoted Tweet 💬 2 🔄 1 ❤️ 18 👀 3680 📊 3 ⚡