Persistent Pre-Training Poisoning of LLMs

257 views

About this video

Advanced · Hand-picked by a BrighterTube curator · found via round-3 back catalogue of a channel Lumen picks video by video · 82% confidence

Studies how poisoned pretraining data persists and affects downstream LLM behavior across fine-tuning. Teaches data poisoning attack mechanisms in modern language model training.

A 33-minute video from Google TechTalks: “Persistent Pre-Training Poisoning of LLMs”.

Video by Google TechTalks, played through YouTube's embedded player.

Part of Differential Privacy for ML