Meta Drops New Model, Mythos, RoboLamp | Luther Lowe, Dan Primack, Lior Susan, Feross Aboukhadijeh, Qasim Mithani, Jaleh Rezaei, Jeremy Philip Galen

Overview

This episode dives into the latest advancements in AI and cybersecurity, featuring discussions on Meta's new AI model Muse Spark, Anthropic's Mythos, and the broader implications of AI in cybersecurity and business. Guests also explore the challenges of app store monopolies, the rise of prediction markets, and innovations in physical industries like chip design and manufacturing.

Notable Quotes

- In an AI-powered world, all phishing becomes spear phishing. – Jeremy Philip Galen, on the evolving sophistication of cyberattacks.

- The reason SpaceX can have that type of IPO is because they solve something that is really, really hard. – Lior Susan, on the value of solving complex problems in physical industries.

- The Apple App Store is basically like the worst DMV in the world. – Luther Lowe, on the restrictive nature of app store ecosystems.

🧠 Meta's Muse Spark and the AI Race

- Meta launched Muse Spark, a closed-source large language model designed to power its AI chatbot and features across its platforms.

- John Coogan highlighted Meta's shift from open-source to closed models, citing cost and safety concerns as driving factors.

- The model's performance benchmarks show it competing with OpenAI, Anthropic, and Google, though it underperforms in some areas like Arc AGI 2.

- Meta's focus on proprietary data and efficient compute usage positions it as a serious contender in the AI space.

🔒 Anthropic's Mythos and Cybersecurity Implications

- Anthropic's Mythos model excels at finding zero-day vulnerabilities, with applications in cybersecurity for critical infrastructure.

- The model is currently restricted to select partners like Apple, Google, and JPMorgan Chase to prevent misuse.

- Dan Primack noted skepticism around Anthropic's marketing strategy, which emphasizes the model's potential dangers while limiting access.

- Mythos underscores the growing role of AI in preemptively addressing cybersecurity threats.

⚙️ Innovations in Physical Industries

- Lior Susan of Eclipse Ventures discussed investments in companies like Cerebras (wafer-scale chips) and VulcanForms (high-precision metal manufacturing).

- Cerebras' wafer-scale integration addresses the limits of Moore's Law, while VulcanForms uses 160-laser systems to revolutionize metal part production.

- Susan emphasized disciplined company-building in capital-intensive sectors and predicted a strong future for real asset companies in public markets.

🛡️ Cybersecurity and Supply Chain Attacks

- Feross Aboukhadijeh of Socket detailed how his platform detected a malicious update to the Axios npm package within six minutes.

- The attack, orchestrated by North Korean actors, involved sophisticated social engineering to compromise the maintainer's account.

- Socket's system analyzes open-source packages across 19 ecosystems, using AI and static analysis to identify threats.

- The rise of AI agents and vibe coding has increased the attack surface, making supply chain security a top priority for organizations.

📱 App Store Monopolies and Policy Challenges

- Luther Lowe from Y Combinator criticized Apple and Google's control over app distribution, likening the App Store to a DMV.

- He advocated for the BASE Act, which aims to curb anti-competitive practices by dominant platforms.

- Lowe highlighted the need for alternative app stores and sideloading to foster innovation and competition.

- The discussion also touched on the challenges of developing and distributing vibe-coded apps under current app store policies.