Open Models Don't Have to Run on Clouds

Dario Amodei argues that open models aren't really open, because in the end you still have to host them on a cloud to run inference. Goju stops the clip and pushes back: that framing only holds if every open model is a large language model that needs a full cloud instance. It isn't. The whole point of small language models is that a 3B or 8B model downloads and runs locally on a high-end laptop, so inference is free and the cloud is optional. The disconnect, Goju argues, is an LLM-only worldview treating its own assumptions as the whole map. The segment opens on a related thread with Mac about the claim that people are shipping 30,000 lines of code a week. Goju separates throwaway MVPs and prototypes from a single complex system you have to maintain. At 30,000 lines a week you'd reach 1.5 million lines in a year, and no current AI system can reason about a footprint that large without burning its entire context just reading the code. Small, maintainable footprints beat volume over time. The takeaway is the same in both halves: match the tool to the actual task instead of defaulting to the biggest model or the highest line count. Open weights that run locally change the economics, and code you can maintain beats code you can only generate. Counter-arguments are welcome in the comments and the Discord. 🔍 Topics covered: - Why "open models must run on clouds" only holds for LLMs - Small language models running locally on a laptop - 3B and 8B open single-node models and free local inference - Open weights versus open source, and why the distinction matters - The 30,000-lines-a-week claim and MVPs versus real systems - The context death spiral on very large codebases - Matching the tool to the task instead of defaulting to the biggest model 💬 Do you run any models locally today, or does everything you use still go through a cloud? 🔔 Subscribe for no-hype tech analysis: https://youtube.com/@gojutechtalk 📺 Related: The Dangers of Over-Reliance on LLMs and AI https://youtu.be/U1Dhfij4Uy0 📺 Related: The Problem With Today's AI (In Simple Language) https://youtu.be/Cl7x2OhbPwU 📺 Related: Good Software Development & the Future of AI for Code (In Plain Language) https://youtu.be/JR22KE6kLMA #SLM #SmallLanguageModels #LocalAI #OpenWeights #OpenSource #Inference #AICoding #GojuTechTalk