Today's Top Episodes
#262 Aliia Roza - Russian Sex Spy on Seduction Perfumes, Sexpionage and Honeypot Tradecraft
Ever wondered how real-life spies operate? A former Russian operative spills the secrets of seduction, espionage, and the dangerous game of honeypots.
Jailbreaking AGI: Pliny the Liberator & John V on AI Red Teaming, BT6, and the Future of AI Security
Is AI security 'security theater'? Dive into the high-stakes world of AI jailbreaking and red teaming, where the future of intelligent systems hangs in the balance.
Everything is gambling now
Your news feed is becoming a casino. Explore how 'prediction markets' are turning everything into a bet and what it means for our future.
#2426 - Cameron Hanes & Adam Greentree
Beyond the hunt: Discover how managing wild predators and embracing extreme physical struggle can unlock ecological balance and profound personal growth.
AI Bubble, Inflation, and the Limits of Monetary Policy | Jason Furman
Is the AI boom a bubble waiting to burst? Unpack the forces driving inflation and the limits of central banks to save us from an affordability crisis.
Viewing Podcast: Podcast
AI
Arts
Business
Crypto
Finance
Health
History
Interviews
Investing
Macro
Misc
News
Politics
Product
Programming
Science
Social
Startups
Technology
VC
Jailbreaking AGI: Pliny the Liberator & John V on AI Red Teaming, BT6, and the Future of AI Security
December 16, 2025
- Jailbreaking AI models is a complex "cat and mouse game" where attackers exploit the expanding attack surface to circumvent guardrails, often at the expense of model capability and creativity.
- The current focus on "security theater" and model lobotomization through excessive guardrails is viewed as ineffective, with a call for more collaborative approaches like open-sourcing data to advance collective understanding.
- AI red teaming extends beyond model vulnerabilities to encompass the entire technology stack, aiming to protect both models from bad actors and the public from rogue models.
After LLMs: Spatial Intelligence and World Models — Fei-Fei Li & Justin Johnson, World Labs
November 25, 2025
- World tokenization and representation are key challenges for world models, with ongoing debate on whether to focus on explicit physical properties or latent emergent understanding.
- Spatial intelligence is presented as a distinct and crucial frontier for AI, complementing linguistic intelligence by focusing on understanding, reasoning, and interacting within 3D space.
- The development of world models is discussed in the context of compute scaling and the evolving role of academia versus industry in AI research, highlighting the need for resourcing academic AI work.
Anthropic, Glean & OpenRouter: How AI Moats Are Built with Deedy Das of Menlo Ventures
November 14, 2025
- Anthropic's rapid growth and market position: Anthropic has achieved exceptional growth, potentially being the fastest-growing software company, now reaching significant revenue milestones and challenging established players like OpenAI in the enterprise LLM API market.
- The complexity of enterprise search and AI adoption: Building effective enterprise search is challenging due to data volume, query diversity, and the need for different evaluation techniques than consumer search, with AI accelerating rather than creating this market.
- The impact of AI on software engineering and developer skills: The increasing reliance on AI coding assistants raises concerns about the erosion of fundamental problem-solving skills, the potential for new attack vectors, and how future developers will be trained.
How Frontier AI + Virtual Biology Can Help Us Cure All Diseases
November 6, 2025
- CZI is shifting its primary focus to basic science, particularly in the intersection of AI and biology, aiming to develop foundational tools and data to accelerate scientific discovery rather than translational work.
- The Biohub model emphasizes interdisciplinary collaboration and institution building, bringing together scientists and engineers from different fields and institutions to tackle complex biological problems.
- The future of medicine is envisioned as N-of-1 precision medicine, leveraging AI and advanced biological data to understand individual genetics and develop personalized treatments for both rare and common diseases.
How Zyphra went all-in on AMD + Why Devs feel faster with AI but are slower — with Quentin Anthony
November 3, 2025
- Zephra is investing heavily in the AMD ecosystem for model training, including porting Flash Attention 2 and finding that MI300X can outperform H100 for certain workloads due to its memory bandwidth and VRAM.
- The speaker emphasizes the importance of custom kernel development and understanding hardware specifics for optimizing AI models, rather than solely relying on higher-level frameworks like Triton or Mojo, especially when targeting specific hardware like AMD's MI300X.
- Open-source efforts in AI research, like those at Luther AI, should focus on interpretability and specific, well-funded projects rather than broad, large-scale model development, aiming to provide valuable insights and replicable research for the community.
The Agents Economy Backbone - with Emily Glassberg Sands, Head of Data & AI at Stripe
October 30, 2025
- Stripe is developing agentic commerce protocols in partnership with OpenAI to enable businesses to interact with AI agents for buying and selling.
- The company is addressing new forms of fraud, particularly "friendly fraud" and abuse of free trials and refunds, which pose existential threats to AI businesses due to high inference costs.
- Stripe is innovating in flexible billing models, such as token billing and outcome-based pricing, to help AI businesses manage fluctuating LLM costs and align pricing with value.
Context Engineering for Agents - Lance Martin, LangChain
September 11, 2025
- Context engineering has emerged as a key challenge in building effective agents, focusing on managing and optimizing the flow of information from various sources, including tool calls, into the language model's context window.
- Offloading context, summarizing tool outputs with high recall to minimize the context window, and isolating context for different agents, represent key techniques under consideration to overcome the challenges of naive agent architectures and make the most of the resources available.
- The "bitter lesson" also applies to AI engineering, emphasizing that while specific structures and assumptions might be necessary for initial success, teams must continue to assess and remove bottlenecks to leverage exponentially improving models effectively.
History of Generative Media with Fal.ai
September 5, 2025
- FAL's pivot to focusing on generative media models, specifically image and video, was driven by the insight that it was a faster-growing niche market with less competition compared to language models dominated by tech giants.
- FAL has achieved significant performance gains, up to 10x faster inference, by building a specialized inference engine with custom kernels optimized for various diffusion models and GPU architectures.
- Video models are now a major driver of revenue for FAL, with the company actively partnering with closed-source model developers to optimize and distribute their models, reflecting the increasing importance of video generation in the generative media landscape.
Long Live Context Engineering - with Jeff Huber of Chroma
August 19, 2025
- Chroma's core mission is to help developers build production-ready AI applications by making the process more like engineering and less like alchemy.
- Context engineering, the process of strategically managing the information fed into language models, is critical for AI startups and developers, as models suffer from "context rot" when processing excessive or irrelevant data.
- Chroma is focused on empowering developers by improving developer experience through things like a zero-config cloud platform, which makes building AI applications easier and more cost effective, and investing in open-source tools and resources like generative benchmarking.
The AI Agenda: GPT5 leaks and the business of AI News — Steph Palazzolo, The Information
August 6, 2025
- The Information's AI reporter Steph Palazolo describes her role as similar to a VC, spending most of her time networking and gathering insights from founders, researchers, and investors to stay ahead of mainstream AI news.
- Palazolo shared her perspective on the AI inference market, highlighting skepticism around GPU resellers versus the potential for these companies to scale with the growth of AI applications.
- Palazolo discussed the challenges of covering AI labs like Meta and OpenAI, including navigating PR strategies, interpreting "breaking news," and the ethical implications of AI development and usage.