What’s getting tongues wagging in AI security? A joint study by Anthropic, the UK AI Security Institute and the Alan Turing Institute shows that just 250 poisoned documents can implant a “backdoor” in any large language model—whether it’s a 600 M or a 13 B parameter beast—meaning tiny injections of malicious data can compromise even huge models.
On a lighter note, Diwali’s here and you can snag 20% off all live courses with coupon code AI20 at Krishnaik.in. Check out their Ultimate RAG Bootcamp, enroll via the links, or ring +91 91115 33440 / +91 84848 37781 if you’ve got questions.
Watch on YouTube
Top comments (0)