Researchers from Anthropic, the UK AI Security Institute and the Alan Turing Institute found that just 250 carefully crafted “poisoned” documents are enough to implant a backdoor in a large language model—no matter if it’s a 600 million-parameter model or a 13 billion-parameter behemoth.
On the festive front, Krish Naik is celebrating Diwali with a 20% discount on all live courses. Use coupon code AI20 at checkout (https://www.krishnaik.in/liveclasses or the Ultimate RAG Bootcamp link) and reach out to +91 91115 33440 or +91 84848 37781 for any questions.
Watch on YouTube
Top comments (0)