After Things Podcast

The episode opens with a discussion of rapid recent AI releases and whether AI has "hit a wall." Andrew points to OpenAI's O3 and Google video models as evidence that capabilities are still advancing, while Justin uses the ARC Prize and AGI as the lens for asking how quickly systems are improving and whether a reasonable AGI label could arrive within the next year. Andrew's response emphasizes the "jagged frontier": models can be very strong on some tasks and weak on others, so benchmark gains do not translate cleanly into broad intelligence. A major middle section focuses on DeepSeek, which the hosts describe as a highly capable Chinese model that has excited open-source enthusiasts and alarmed frontier-lab skeptics. Andrew argues the model should be understood in context: export restrictions may have pushed efficiency work, but the model likely also benefited from distilled outputs from frontier models and other structured training data, so it is not a clean from-scratch achievement. The episode then turns to YouTube's AI-training opt-in controls, the copyright and compensation questions around creator data, the growing reputational stigma around obvious AI-generated creative work, and predictions for 2025 that include more AI-automated workflows, a company announcing AGI, and more AI-assisted email handling. Key topics AI benchmark gains and the "hitting a wall" narrative: The hosts contrast media claims that AI has stalled with examples of large benchmark jumps, especiall

In this episode, Andrew Mayne, Justin Robert Young, and Brian Brushwood take listeners on a fascinating exploration of entertainment’s past, present, and future. They kick off with a nostalgic look at how theme parks like Universal Studios have evolved, highlighting the technological advancements in attractions such as the Born Stunt Spectacular. The conversation then shifts to the potential of AI in creating immersive experiences, with Andrew sharing insights into reasoning models and the concept of AI agents. They also touch upon the importance of classic sci-fi literature, like ‘A Canticle for Leibowitz’, in understanding the roots of many modern narratives. The episode wraps up with a discussion on the implications of AI in programming and the potential for AI to revolutionize how we interact with technology and each other. Picks: Justin Robert Young: DEF CON Plan about Audio and Electronics Brian Brushwood: A Canticle for Leibowitz Andrew Mayne: Nosferatu by Robert Eggers

The episode opens with a discussion of OpenAI's Shipmas announcements and a comparison with Google's recent AI releases. The hosts focus on OpenAI's o3 model, describing it as a real, usable research milestone and noting that it scored highly on the ARC Prize benchmark and coding evaluations, while also acknowledging that some announced features are not immediately available to everyone. The conversation then broadens into how current AI tools are being used in practice. The hosts talk about ChatGPT integrations with Notion, desktop and screen-sharing features, model switching when one tool is not suited to a task, and the brittleness of AI outputs when prompts or settings change. The latter half shifts to robotics and simulators, especially how physics simulation could accelerate robotics development and how cheaper, more capable robots could change labor and local production. The episode closes with several media picks. Key topics OpenAI Shipmas and Google's AI announcements: The hosts compare OpenAI's 12 days of releases with Google's announcements, arguing that OpenAI's products are more immediately usable while Google's demos appear more limited in access. o3, ARC Prize, and AI benchmarks: A major segment focuses on o3, the ARC Prize visual reasoning benchmark, and coding evals. Andrew explains how the benchmark works and why the o-series models' higher scores matter. AI workflow integration and desktop assistance: The speakers discuss ChatGPT working with Notion, screen

The episode opens with the hosts talking about new live multimodal AI features in ChatGPT and Google Gemini, including Andrew's demo of showing ChatGPT a card trick over live video. They note that these features had been demonstrated earlier and are now shipping, but emphasize that backend compute, server connections, and GPU supply make rollout slower than some people expect. Most of the episode is spent on OpenAI's Sora and other video generators. The hosts discuss how to use Sora, including starting from a strong image or uploaded video, using storyboards, keeping generations short, trying lower resolutions first, using remix tools, and learning from the featured/recent feeds. They repeatedly stress current limitations in physical reasoning, object relationships, and variable binding, while also praising Sora for b-roll, companion footage, character coherence, and other creative uses. The episode closes with a short TV-picks segment covering Lower Decks, Strange New Worlds, Foundation, and Skeleton Crew. Key topics Infrastructure limits behind live AI features: Andrew says live video and multimodal AI depend on significant backend compute and connections, so they are not just app updates and cannot be scaled instantly. AI as a supportive audience member: In the live card-trick demo, the hosts are struck by how ChatGPT behaves like an encouraging spectator, staying supportive rather than immediately correcting the trick. Using prompts, storyboards, and reference images in S

In this episode, Andrew Mayne, Justin Robert Young, and Brian Brushwood kick things off with a brief chat about the latest addition to Justin’s family and the implications of raising a child in today’s tech-saturated world. The conversation quickly shifts to Elon Musk’s recent showcase of Tesla’s advancements in robotics, including the Optimus robot and the Cyber Cab. The hosts speculate on the impact of these technologies on the future, from personal robotics to autonomous transportation networks. Additionally, they touch upon the phenomenon of sky quakes, debating their possible explanations and expressing skepticism about their origins. Throughout, the trio maintains a slightly irreverent tone, mixing in personal anecdotes and broader societal observations. Picks: Brian Brushwood: Alchemy: The Dark Art and Curious Science of Creating Magic in Brands, Business, and Life by Rory Sutherland Justin Robert Young: The Peripheral (TV Show) Andrew Mayne: The 13th Warrior (Movie)

The episode opens with a long discussion of SpaceX successfully catching the Starship booster with Mechazilla. The hosts focus on the scale of the tower and booster, the surprise and delight of the SpaceX team, and what the feat implies for fully reusable rockets. They also broaden the conversation into Elon Musk's impact, conviction and persistence in engineering, and how institutions and experts can be wrong about what is possible. [L21-L29, L33-L41, L47-L57] The middle of the episode turns to robotics and AI. The hosts discuss Tesla's Optimus robots at the We, Robot event, including the gap between what was demonstrated and what was actually autonomous. They then spend a long stretch on an Apple paper about reasoning benchmarks, arguing that a small prompt change can dramatically improve performance and that the paper overstates the case against AI reasoning. The back half becomes a hands-on demo of local AI and Ollama, plus creative prompting tests, before ending with picks for Ollama, The Apprentice, Tribalism is Dumb, and Civil War. [L65-L85, L139-L157, L195-L205, L221-L237, L375-L445, L471-L493] Across the AI discussion, Andrew argues that model capabilities are improving quickly and that skepticism often comes from narrow benchmarks, outdated assumptions, or prior investments in other approaches. Brian shifts toward a pragmatic stance that AI use is mainly a productivity issue and that people care more about the output than the method. The episode closes with a shared

The episode opens with Andrew detailing OpenAI's Dev Day announcements, especially the real-time API for continuous text or audio conversations and demos aimed at customer support and phone ordering. The hosts then debate AI as a replacement or augmentation for customer service, with Brian and Justin emphasizing how frustrating human support can be and how useful a capable AI agent might be if it can actually solve problems. Andrew walks through several additional OpenAI features, including prompt caching, easier fine-tuning, memory controls in ChatGPT, and the new Canvas document-editing mode. The conversation also covers a study comparing doctors and GPT-4 on diagnostic tasks, Meta's new video model and the Movie Gen demos, and then shifts into picks where Justin praises Agatha All Along, Brian promotes Achewillow, and Andrew recommends Only Murders in the Building. Key topics Real-time conversational APIs for voice and text: Andrew explains OpenAI's real-time API for nonstop voice or text conversation and compares it with similar work from other companies like Deepgram. AI as a replacement or augmentation for customer support: The hosts describe AI as potentially better than scripted human support if it can resolve issues quickly on either the calling or receiving end. Superhuman persuasion and AI negotiation risks: Andrew raises the possibility of highly persuasive AI being used for manipulation, while also imagining AI lawyers and negotiators. Prompt caching and cheaper

In this episode, Andrew Mayne, Justin Robert Young, and Brian Brushwood kick things off with a chat about the weather before diving into the world of augmented reality and virtual reality. They discuss the limitations of Apple’s Vision Pro and the potential of Facebook’s Project Orion, comparing the two and expressing their hopes for the future of AR. The conversation then shifts to acoustic holograms, showcasing how sound can be used to manipulate objects in space, creating what could be the future of holographic displays. The trio also touches on the advancements in AI, demonstrating how smaller, faster models can generate content and perform tasks with impressive speed and accuracy. Throughout the episode, the hosts maintain a slightly irreverent tone, mixing in their personal experiences and opinions on the tech industry. Picks: Brian Brushwood: I’m Beginning to Get Worried About This Black Box of Doom by Jason Pargin Justin Robert Young: Mr. McMahon on Netflix Andrew Mayne: Alien Rom

The episode opens with the hosts joking about wanting burritos immediately and turns into a discussion of faster delivery systems. Andrew introduces Zipline's drone-delivery model, describing how it keeps the aircraft high above the ground and lowers cargo by line to avoid noise and landing-safety problems. That leads naturally into a broader conversation about autonomous transport, including Waymo's route-based ride service and Zoox's purpose-built vehicle design, along with speculation about future mobile rooms, containers, and other vehicle-as-space ideas. The middle of the episode moves through AI and brain-interface ideas. The hosts discuss an AI-only social network, using ChatGPT as a vocabulary and etymology aid, and then a playful cave-packing exercise that leads into a real cave story about Cheetos left in Carlsbad Cavern and the microbial ecosystem it supported. From there they broaden into Mars ethics, robotic exploration, AI tool-building and sunk-cost thinking, before ending with a long entertainment segment on The Wire, The Sopranos, and Beetlejuice Beetlejuice, plus praise for Michael Keaton. Key topics Drone delivery designed to stay aloft: Andrew describes Zipline as a delivery system that drops cargo from a drone hanging high in the air, avoiding the safety and noise issues of landing on a porch or sidewalk. Autonomous vehicle design tradeoffs: Waymo is discussed as a mapped driverless ride service, while Zoox is framed as a boxier, purpose-built autonomous

The episode opens with a long discussion of OpenAI's Strawberry / O1-style reasoning models. Andrew Mayne explains that these models seem to work better when asked to break problems into steps, use tools, and reason through tasks in a more structured way than ordinary one-shot chat models. The hosts compare this to prompt engineering, discuss examples like decimal comparisons and counting the R's in "strawberry," and talk about how longer structured prompts, patience, and using the right model for the right task can improve results. Later, the conversation broadens into AI evaluations, benchmark gaming, model stacking, tool use, and concerns about AI persuasion. Andrew argues that leaderboard results can be misleading and that models often look strong in short tests but deteriorate with longer contexts, while Justin notes that eval methods themselves are still immature. They also discuss a Science paper about GPT-4 Turbo persuading people away from conspiracy beliefs, which Andrew frames as manipulative and alarming. The episode then moves into a playful Matrix screening story, a discussion of Polaris Dawn and private spacewalking, and the show ends with Netflix media picks. Key topics Reasoning models as step-by-step task solvers: Andrew describes Strawberry / O1 as a model that performs best on long, detailed, multi-step tasks, especially when asked to break work into steps and think through a problem. Prompt engineering for better outputs: The hosts discuss writing longer