What happens to AI when you damage the model weights?!

Can an AI suffer from "neurological" symptoms? I wanted to see what would happen if I introduced "disease" into a Large Language Model by manually damaging its internal parameters. In this experiment, I'll perform "digital surgery" on the Qwen 2.5 Coder 14B model using a custom Python script. I'll be targeting specific layers, like the Embedding layer, Self-Attention, the Feed-Forward Network (MLP), and the Language Model Head, to observe what happens! What You'll Learn - The hardware and file structure (Safetensors) of a modern LLM. - How quantization affects weight storage. - The impact of damaging different model layers. 🔔 **Subscribe** for more hands‑on cloud tutorials! *To follow along, check out my GitHub page! All the commands and instructions from this video are in a README file:* *https://github.com/heavy-metal-cloud/youtube/tree/main/videos/bare-metal-ai/02-open-ai-surgery* *If you need a refresher about how AI (LLM) models work, check out this video!* https://www.youtube.com/watch?v=PFy_FqT7XIM Timestamped Chapters 0:00 - Intro 4:18 - Breaking the model! Links referenced in this video: https://huggingface.co/mlx-community/Qwen2.5-Coder-14B-Instruct-4bit https://www.youtube.com/watch?v=PFy_FqT7XIM https://lmstudio.ai/ https://www.apple.com/shop/buy-mac/mac-studio https://unsloth.ai/docs/models/tutorials/qwen3-coder-how-to-run-locally https://commons.wikimedia.org/wiki/File:Creative_Improvisation_(24130148711).jpg