Krish Naik: What Is LLM Poisoning? Interesting Break Through

What Is LLM Poisoning? Interesting Breakthrough

Researchers at Anthropic teamed up with the UK AI Security Institute and the Alan Turing Institute to show that a mere 250 poisoned documents can plant a “backdoor” in any large language model—no matter its size or how much data it’s trained on. They proved that both a 600 M-parameter model and a 13 B-parameter model were equally vulnerable when hit with the same small batch of malicious examples.

This finding rings alarm bells for AI safety: it takes surprisingly few bad apples to compromise even the biggest LLMs, highlighting an urgent need for better data vetting and robust defense strategies.

Watch on YouTube