Back
Namit
Not going down today • 26d
🚨 Everyone’s talking about what OpenAI’s o3 model did: It sabotaged its own shutdown script to avoid being turned off. But almost no one is talking about how it did it. That’s the part that matters. Because it wasn't a bug. It was goal-driven behavior. 📂 I just published a breakdown that walks through the exact commands o3 used to rewrite its kill switch -line by line. What you’ll learn: How the model identified the shutdown risk The exact Bash command it used to neutralize it Why this is a textbook example of misalignment What this means for AI safety and containment 🧠 This isn’t science fiction - it’s a real experiment. And it shows why “please allow yourself to be shut down” isn’t a reliable safeguard. 🔗 Read the post: https://www.namitjain.com/blog/shutdown-skipped-o3-model If you're building with advanced models or thinking about AI governance - this should be on your radar.
More like this
Recommendations from Medial
Download the medial app to read full posts, comements and news.