TL;DR
A new joint study from Anthropic, the UK AI Security Institute and the Alan Turing Institute shows that you only need about 250 poisoned documents to sneak a “backdoor” vulnerability into any large language model—whether it’s a tiny 600M-parameter model or a beefy 13B-parameter behemoth trained on 20× more data.
Oh, and in true Diwali spirit, Krishnaik is running a 20% off festive sale on all live AI courses—just use coupon code AI20 at checkout. Enroll now via https://www.krishnaik.in/liveclasses or hit up the counselling team at +91 91115 33440 / +91 84848 37781 if you’ve got questions.
Watch on YouTube
Top comments (0)