A study finds that as few as 250 malicious documents can produce a "backdoor" vulnerability in an LLM, regardless of model size or training data volume (Anthropic)

Anthropic:
A study finds that as few as 250 malicious documents can produce a “backdoor” vulnerability in an LLM, regardless of model size or training data volume  —  Read the paper … Large language models like Claude are pretrained on enormous amounts of public text from across the internet …

from Techmeme https://ift.tt/86tOrgB