AI Summary
→ WHAT IT COVERS Albert Brotherton (22) and Boris Radilov (24) present Nanogram, a TikTok-style feed of AI-generated mobile games, launched mid-January with 100,000 users. Separately, One X founder Bernd Øverås details Neo, a 66-pound home humanoid robot priced at $20,000, shipping to early adopters in 2026 with a world-model AI architecture. → KEY INSIGHTS - **Nanogram's engagement metrics:** Within six weeks of launch, Nanogram reached 100,000 users with 20% classified as power users who play more than 25 games per session. Average session time runs 21 minutes across two daily sessions, totaling roughly 50 games per day per power user — without any recommendation algorithm in place, suggesting strong organic retention from the scroll-and-play format alone. - **AI game creation pipeline:** Nanogram uses a custom game engine paired with Google Gemini and agentic tool-calling to generate playable 3D games from a single text prompt in 60–90 seconds. Tools include 3D mesh generation, 2D pixel art, and sound creation. Users can then remix their own games or fork other creators' games through unlimited follow-up prompts, lowering the creation barrier to near zero. - **Interactive ad format opportunity:** Brands currently have no scalable interactive ad format across social platforms. Nanogram's model allows brands to build playable game ads — a pizza chain could let users build a custom pizza in-game with a direct order call-to-action. Engagement and conversion rates for interactive content exceed passive video, making this a structurally differentiated ad inventory compared to TikTok or Instagram. - **Neo's world model vs. VLMs:** One X argues that vision-language models (VLMs) process static 2D screenshots without capturing 3D spatial or temporal dynamics, making them insufficient for general physical AI. Their world model trains robots to simulate forward consequences of actions — similar to human mental simulation — enabling safer task planning. This architecture allows Neo to handle novel voice-commanded tasks not present in training data. - **10,000 robots equals YouTube's upload rate:** One X estimates that 10,000 deployed Neo units generate robot training data at roughly the same rate as YouTube's current upload volume. Unlike video data, robot data captures three components — internal goal state, chosen action, and observed result — making it significantly richer per token than web-scraped video, potentially requiring less total data to reach general intelligence benchmarks. - **Neo's physical design reduces manufacturing cost at scale:** Neo weighs 66 pounds — roughly one-third of competing humanoid robots — directly cutting raw material requirements at billion-unit scale. One X uses tendon-pull motors instead of classical gears, enabling loose manufacturing tolerances, fewer parts, and no rare materials beyond magnets. Robots are already assembling robots in their Hayward facility, with capacity for tens of thousands of units annually. → NOTABLE MOMENT One X's founder revealed he has kept a home robot for multiple years — first with the industrial Eve model, now with Neo. He described asking Neo to locate and read a Post-it note on a wall, a task absent from training data, and the robot completed it successfully, though it failed to repeat the same task on a second attempt. 💼 SPONSORS [{"name": "Render", "url": "https://render.com/twist"}, {"name": "Every", "url": "https://every.io"}, {"name": "LinkedIn Jobs", "url": "https://linkedin.com/twist"}] 🏷️ AI Game Generation, Humanoid Robotics, Short-Form Gaming, World Models, Creator Monetization, Interactive Advertising