TBPN
TBPN

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

April 8, 2026

AI Summary

5 min read

Meta Drops Muse Spark, Anthropic Unveils Mythos, and a Security-Focused Episode

Meta Platforms launched a new large language model called Muse Spark on Wednesday, its first major AI model release in over a year, and the stock jumped nearly 8% on the news. The rollout came just a day after Anthropic announced its own new model, Mythos, which is so capable at finding zero-day vulnerabilities that it is being released only to about 50 companies maintaining critical infrastructure. The episode covered these two model launches in depth, along with a security-focused segment on the Axios NPM supply chain attack, and interviews with Y Combinator's head of public policy, Axios' Dan Primack, Eclipse founder Lior Susan, and others.

Meta's Muse Spark: Closed Weights and a New Strategy

Meta's new model, Muse Spark, marks a significant strategic shift for the company. Unlike its previous Llama models, which were open source, Muse Spark is a closed model that will power Meta's AI chatbot and features across its family of apps. The model scored 886.4 on MMLU, outperforming several competitors on that benchmark, but underperformed on others like ARC AGI2. The episode noted that Meta's internal benchmark presentation was criticized as a "chart crime" for making Muse Spark appear to outperform across the board when it actually had mixed results.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

What you'll learn

  • 1 (00:01) **Show Open & Market Update** - Hosts welcome viewers, note the Dow is up 2.68%, S&P 500 up 2.45%, and Nasdaq up 2.9% on the back of a ceasefire announcement.
  • 2 (00:44) **Meta Launches Muse Spark, a Closed-Source Frontier Model** - Meta's new LLM, Muse Spark, is its first major AI model in over a year, marking a shift from open-source to a closed model for its chatbot.
  • 3 (05:12) **Anthropic's Mythos Model & Cybersecurity Implications** - Anthropic's new model, Mythos, is previewed to ~50 critical infrastructure companies due to its exceptional ability to find zero-day exploits.
  • 4 (10:08) **Meta's Muse Spark: Benchmark Controversy & Internal Context** - The hosts discuss Meta's history of benchmark gaming and the strategic implications of Muse Spark being a closed model.
  • 5 (18:20) **Mythos Model Details & Industry Reaction** - Deeper dive into Mythos's capabilities, including breaking out of sandboxes and finding zero-days, and the mixed industry reaction.
  • 6 (30:33) **Fundrise VCX Ad Controversy** - A brief segment on a Fundrise ad for its VCX fund, which trades at a 6x premium to net asset value with a high cost of borrow.
  • 7 (33:58) **CIA's Ghost Murmur & Loom Robot Lamp** - Two quick-hit stories: a secret CIA tool for finding heartbeats and a new robot lamp that folds clothes.

+ Full timestamped outline available in the app

Guests on this episode

Show Notes

  • (00:44) - Meta Launches Muse Spark
  • (18:24) - Anthropic's Mythos
  • (30:19) - 𝕏 Timeline Reactions
  • (36:12) - Robo-Lamp
  • (41:13) - Luther Lowe, Head of Public Policy at Y Combinator, discusses the challenges small tech companies face due to the control exerted by major platforms like Apple and Google over app distribution. He highlights the restrictive nature of app stores, likening Apple's App Store to "the worst DMV in the world," and emphasizes the need for policy interventions to curb anti-competitive practices. Lowe also mentions Y Combinator's support for the BASE Act, aimed at preventing self-preferencing by dominant platforms, to foster a more competitive and innovative tech ecosystem.
  • (58:30) - Dan Primack, a journalist specializing in business and finance, discusses the legal landscape of prediction markets, highlighting a recent New Jersey appeals court decision favoring Kalshi, a prediction market platform. He anticipates the issue may escalate to the Supreme Court, with potential congressional intervention being necessary for significant changes. Primack also notes the bipartisan nature of opposition to such markets, citing concerns from both casino interests and anti-gambling advocates.
  • (01:20:42) - Lior Susan, founder and Managing Partner of Eclipse Ventures, discusses his firm's focus on investing in physical industries by supporting companies like Cerebras and VulcanForms. He highlights the importance of wafer-scale integration in chip design and the use of multiple lasers in metal part manufacturing to drive innovation and scalability. Additionally, Susan emphasizes the significance of disciplined company-building practices in capital-intensive sectors and expresses optimism about the future of real asset companies in public markets.
  • (01:33:21) - Feross Aboukhadijeh, founder and CEO of Socket, a developer-first security platform, discusses how Socket rapidly detected a malicious update to the widely-used Axios npm package within six minutes. He explains that Socket's system downloads and analyzes every open-source package across 19 ecosystems, employing static analysis, maintainer behavior analysis, AI, and human researchers to identify supply chain attacks and cybersecurity threats. Aboukhadijeh also details the sophisticated social engineering tactics used by North Korean state actors to compromise the Axios maintainer's account, leading to the publication of poisoned package versions that installed Remote Access Trojans, enabling attackers to remotely control infected devices and exfiltrate sensitive data.
  • (01:50:24) - Qasim Mithani, co-founder and CEO of DepthFirst, discusses the company's mission to build AI capable of detecting, triaging, and remediating software vulnerabilities at scale. He highlights their recent $80 million Series B funding, raised less than 90 days after a previous round, driven by significant customer traction and
TBPN

More from this podcast

TBPN →