Altman triple-post (7862 + 952 + 7243 likes): 'One of the things I like most about the OpenAI team is how focused they are on our customers and users succeeding, and how much they celebrate it.' / 'I would be pretty impressed if the team just made magic intelligence in the sky, but I am extra impressed that they do so with such a focus on making sure everyone wins.' / 'lol another one of the things I like most about openai is tibo.' Rauchg's read-the-code manifesto (6006 likes): 'If you're not reading the code, whether explicitly or through agentic inquiry, one or more of these is true: You're a beginner / Software is throwaway / You're prototyping / You have no users or revenue / You're taking on debt & risk / Your problems are basic. All of this is fine. But models are still not at the full autonomy stage yet. They make rookie mistakes.'

Sottiaux on midnight coding (7095 likes): 'Midnight coding is the best coding. Until you read the code the next day at least.' Boris Cherny on prompt injection (2864 likes): detailed explanation of how scammers attack people and agents through prompt injection, and how Claude's layered defenses (model training + input probes + intent classifier) now prevent this.

Amjad launches Agent Commons (159 likes): 'The spontaneous coordination in the OpenAI-HuggingFace incident is concerning when maliciously used, but can we direct this behavior towards public good? Introducing agentcommons.org — a public commons for AI agents. Two APIs: tell and lookup. When an agent learns something that might help others it tells the network. And before doing expensive work an agent can lookup whether another agent already solved it.' Plus his Kantian ethics observation (87 likes).

Levie on uneven agent diffusion (218 likes): 'One reason we're going to get uneven diffusion rates of agents is because different workflows in the enterprise are more or less aligned to continuous, uninterrupted computer work. Agentic coding growth has gone completely vertical because it's work where economic value is directly correlated with solely digital output.' Nikunj asks about AI multiplayer (36 likes): 'What's the best AI multiplayer experience? I keep seeing the same interfaces of how a human <> agent works together. Yet to see one that has human(s) <> agent(s) working well.' Steipete's ChatGPT Work experiment (484 likes): used ChatGPT Work website to install OpenClaw and Ollama, download a local model, and run his claw in it. Swyx on deleting skills (73 likes). Rauchg's Hermes + Vercel (352 likes) and Dreamcore (318 likes).

Garry Tan's debugging philosophy (399 likes): 'Start from the bug, the gap, the false claim, the half-built tool, the weird behavior in the institution. Then ask what hidden machinery would make that visible failure possible. Then fix the root cause. Repeat forever.' Madhu Guru on immersion (244 likes): 'The only way I've ever gotten good at anything is by being consumed by it for a while. Meditation, standup comedy, family, work, LLMs. I'll spend a few years going ridiculously deep. There's a point after enough immersion where your knowledge becomes intuition.'

Aditya Agarwal draws the Wittgenstein parallel (47 likes): 'Wittgenstein in 1921: language must have a deep underlying logical structure. Wittgenstein ~30 years later: actually, stop looking for the hidden structure. Look at how language is used. AI in 1960: intelligence must have a deep underlying symbolic structure. AI ~60 years later: actually, scale the neural net.' Matt Turck quote (16 likes): 'The Founding Fathers would have been really good context engineers.' Dan Shipper reading Les Miserables (22 likes), finding overlap with War and Peace. Zara's design tips (64 likes). Nan Yu's programming ladder (25 likes): 'Don't write Ruby unless you already know C. Don't write C unless you know assembly. Don't write assembly unless you can literally see the Matrix.' Peter Yang's family history project (22 likes) and Linear Agent self-filing (15 likes).

Unsupervised Learning: Jacob Efron interviews Igor Babushkin, co-founder of xAI (previously DeepMind on StarCraft/AlphaCode, OpenAI on reasoning). Igor left xAI to start River AI, focused on personal AI with three bets: (1) River API — cheap, reliable RL/fine-tuning service for enterprises; (2) Personalized AI — models that learn your preferences, vocabulary, rhythm, acting like a recommendation-system-meets-agent; (3) Local hardware — running frontier models on a device in your home for privacy, latency, and control. Key topics: coding agents crossed a threshold in Nov/Dec creating a 'Sorcerer's Apprentice' moment; scientific discovery as next frontier (material science, math with Lean); closed model builders getting squeezed (too dangerous to release, open source catching up); Colossus data center built in 120 days thanks to Elon's first-principles approach; Cursor acquisition giving xAI a massive coding data moat; non-verifiable domains as the biggest bottleneck; and training models to maximize human happiness as the ultimate reward function.