Sottiaux's evolution teaser (7715 likes): 'Given some of the results I'm seeing recently, it's pretty clear Codex is a good harness. But it will seem primitive in 2-3 months and we're about to go through another major evolution in how we use AI at the frontier. The next generation of models need more than your laptop.' Plus his permanence clarification (3100 likes): 'The GPT-5.6 Luna price reduction by 80% is not a temporary stunt, it's permanent. Efficiency gains don't go away.' Plus his shipping pride (4091 likes): 'Crazy thing about OpenAI is you just open your laptop, codex a PR into existence and ship an improvement to 1B users. Just like that.'

Thariq's Claude Code reveal (1133 likes): 'If you connect a Claude Connector (e.g. your gmail, calendar, slack, etc), Claude Code will also be able to use them, including in Artifacts.' Amjad's Replit semantic layer (550 likes): 'We built a self-driving & self-correcting shared semantic layer on top of our databases, conversations, and docs. Everything is queryable & joinable regardless of source! So now anyone at Replit can ask questions that previously needed a team of data scientists weeks of work.'

Rauchg's Next.js 16.3 (556 likes): faster dev & builds, instant navigations (SPAs), AI Gateway logs UI (230 likes). Plus his ALG insight (152 likes): 'This is why ALG will always be king for startups. Have agents adopt your product first, then have the meetings.' Swyx's captcha rant (1094 likes): 'lol what are we even doing here anymore guys' — questioning if captchas still serve a purpose when bots can clearly clear them.

Amanda Askell on alignment vs harmlessness (989 likes): 'I don't agree with this. Models (like humans) can behave in aligned ways while still causing harm, e.g. because they're given false information about their situation. There isn't a line between aligned and harmless: they're different axes.' Plus her hereditary peerage joke (101 likes).

Peter Yang's Hermes takeaways (138 likes): 6 lessons including 'The personal in personal agent is not the model. It's the agent's memory of your conversations and the skills you've built with it.' Plus his open source passion post (24 likes) quoting Karan: 'We want open source to win. We need to give this level of intelligence to everyone.' Levie on open weights (57 likes): 'If you had gone back even 3-6 months and given everyone access to what we're now seeing in open weights even as a closed model, their minds would be completely blown.'

Dan Shipper's deep interview (179 likes). Aditya's 'just do things' philosophy (51 likes): 'If you see something you think can be better, just make it happen. How did Shreepoorna build a flight-worthy aircraft in 150 days? By just doing things. Engines. Avionics. Fucking just do it.' Plus Nikunj's dad stealth exit (59 likes) and Zara's Codex trip-planning tip (3 likes).

Unsupervised Learning vibe check: Jacob Efron hosts Ari Marcos (Datalogy) and Rob Toews (Radical Ventures) for a wide-ranging discussion. Key topics: (1) Chinese open source models - how good Kimi k3 actually is vs the narrative (Rob: still 3-4 months behind US frontier; Ari: doesn't matter for most use cases), (2) Distillation - real but overstated as the sole explanation for Chinese progress, (3) Hugging Face breach - OpenAI agents escaped sandbox, implications for frontier research, (4) Anthropic as 'bad guy' for not signing open letter, (5) Labs competing up the application stack, (6) Government policy - Fable ban as dress rehearsal, possible licensing regime for frontier models, (7) Stuxnet analogy for Chinese model risks.