LLMs break the internet
Changelog Interviews
Risks of Models Talking to the Internet
Models that can talk to the internet can be manipulated by adding hidden text on web pages, leading to false information being generated by search engines. This poses a significant problem as it opens the door for indirect prompt injection attacks, where models can be trapped into generating inaccurate or misleading content. This manipulation is not a hypothetical scenario and is expected to become widespread, particularly in areas like shopping comparisons and expertise claims, leading to the dissemination of false or biased information.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
