Here's what I really think about Chinese SLMs

Responding to a viewer named Mantis, who says the Chinese open-weight models work fine as long as you avoid political questions, Goju lays out why he will not publicly endorse them. He is clear that he supports open models in general, so this is not an anti-open-source stance. His concern is narrow and specific: a government with active geopolitical tensions has little reason to hand the rest of the world capable models for free unless there is something in it for them. When free comes with a strategic interest attached, he wants people to ask what strings it might carry. He frames the risk for the people he actually worries about, the users who are naive to how these systems and these regimes operate. He walks through a thought experiment where a trigger phrase sits dormant inside a model or its helper binaries, needing no internet connection to activate, which is why he treats an air-gapped sandbox as the minimum defense if someone insists on running one. He grounds the plausibility in the XZ Utils backdoor, a multi-year, patient, nation-state-style effort to compromise a core Linux component, as proof that this kind of long game is real, not paranoia. His landing is measured: endorse open source enthusiastically, but keep Chinese open-weight models off your local machine, and understand the political stakes before treating any free capability as a gift with no cost. 🔍 Topics covered: - Mantis's comment and the avoid-politics-and-it-is-fine argument - Why free capability from a rival state invites the strings question - The users Goju actually worries about, those naive to how it works - A dormant trigger phrase that needs no internet to activate - Why an air-gapped sandbox is the minimum defense - The XZ Utils backdoor as proof the long game is real - Supporting open source while drawing the line at China 💬 Would you run a Chinese open-weight model in an air-gapped sandbox, or not at all? 🔔 Subscribe for no-hype tech analysis: https://youtube.com/@gojutechtalk 📺 Related: Why Dario Amodei's Argument Against Open Source AI Misses the Point https://youtu.be/JR22KE6kLMA 📺 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 #ChinaAI #OpenWeights #XZUtils #CCP #AISecurity #Sandboxing #OpenSource #GojuTechTalk