What Is LLM Poisoning?
A new joint study by Anthropic, the UK AI Security Institute and the Alan Turing Institute reveals that injecting just 250 malicious documents into an LLM’s training data can create a “backdoor” vulnerability—no matter the model’s size or the total amount of data it’s trained on. Even a beefy 13B-parameter model is as susceptible as a smaller 600M-parameter one.
On a lighter note, there’s a Diwali deal going on: grab 20% off all live AI courses with code AI20 at Krishnaik’s site. Reach out to their counseling team at +91 91115 33440 or +91 84848 37781 if you’ve got questions.
Watch on YouTube
Top comments (0)