Stop prompting. Start dissecting.
LLMs are no longer just "cool gadgets"—they are critical infrastructure with a "Black Box" problem. From hallucinations to security jailbreaks, we face a reality where we don't truly know how these machines work.
Join GDG on Campus KIIT for a high-intensity workshop on Mechanistic Interpretability. We’re moving beyond the chat box to perform digital surgery on neural networks.
The Mission:
• Identify the Circuits: Find the specific paths responsible for logic vs. hallucinations.
• Visualize the Breach: See how a model decides to bypass safety constraints.
• Steer the Weights: Intervene in internal pathways to control behavior without retraining.
The Playbook:
• The Breach: Live demos of LLM mechanics and jailbreaking.
• Digital Surgery: Colab code-along to reverse-engineer attention heads.
• The Intervention: Hands-on activation steering to stop model "lying".