Chronos

🔥 /goal is currently the MOST powerful feature in Claude Code right now. No more babysitting the agent with endless “continue” every 30 seconds. Just drop /goal + a crystal-clear success condition (“all tests green”, “90% coverage”, “feature fully shipped + documented”) — and Claude runs autonomously until done. Haiku validates it after every turn. The thread breaks it down perfectly: • How it works under the hood • What goals actually work vs what burns tokens • Pro setup (CLAUDE.md + Auto Mode) Game changer for big refactors and full features. Huge thanks to @akshay_pachaar for the breakdown 🔥 Who’s already using /goal in their workflow? Drop your experience below 👇

Just saw @blinktrade reply to @toly with "Kingmade already" 👑 Came out of stealth yesterday and honestly looks pretty promising. Onchain exchange on Solana focused on real speed: verifiable FIFO, physical colocation, and sub-millisecond execution. For people building trading agents or trading perps, this might become proper infrastructure in 2026. Still flying under the radar for most. Curious to see how it develops. Anyone else checking out @blinktrade? #QuantTrading

Leading-edge programmers are 20x more productive than one year ago. Twitter cut 70-80% of staff. platform runs better than before. 33 minutes. free. by Marc Andreessen, General Partner at a16z. → AI vampires: exhausted, euphoric, unstoppable → "builder" replacing coder + PM + designer → AI bloat exposed as corporate layoff cover → golden age thesis: AI as universal superpower This is the closest you'll ever get to seeing inside the mind of Silicon Valley's most powerful venture capitalist. Bookmark & watch. Then start building.

@maarcoofdezz This is the most transparent “inside a hedge fund” talk I’ve seen. The eval pipeline they mentioned is what 99% of retail “AI trading bots” miss. Already watching it twice.

Un ingeniero de IA de Jane Street reveló cómo entrenaron su propio LLM para trading y generar 22,5 mil millones de dólares al año. 16 minutos. Gratis. Directo de quants de primer nivel. Guárdalo y míralo: probablemente sea la charla más honesta jamás publicada sobre “IA dentro de un hedge fund”. Olvídate de los gurús de YouTube vendiendo “bots de trading con IA”. Esta es la visión real desde dentro: datos, entrenamiento, evaluaciones e integración. (Guárdalo antes de que sea tarde)

The two-phase pretraining framework is what caught my attention. Phase 1: general knowledge acquisition Phase 2: reasoning-centric integration The bottleneck was never compute. It was always data ordering. Most people still don't know this.

Stanford just dropped this. Free. From next-token prediction to next-generation intelligence. The future of LLM pretraining — revealed. 48 minutes. free. Stanford University | April 2026. → why next-token prediction hits a ceiling → data ordering strategy that unlocks reasoning → two-phase training framework → what comes after pretraining as we know it This is the closest you'll ever get to seeing inside Stanford's AI research engine. Bookmark & watch. Then start building.

@solana it looks very good