Urgent.News

What's breaking now, across thousands of outlets.

More in Tech

Numbers Do Not Belong in the Model

A missing number gets noticed. A wrong one does not. That single asymmetry is why reading documents with a language model has to be treated differently from writing text with one.

  • Numbers in language models require special attention and treatment.
  • Multi-digit numbers often split into individual units during tokenization.
  • Recommended solution involves extracting numbers from raw document before model involvement.

RAG security: the retrieved document is now your attack surface

Cross-post. Original: stellarbytecapital.com/blog/rag-security Retrieval-augmented generation is the default way to make an LLM useful on your own data: fetch relevant documents, stuff them into the…

  • Retrieval-augmented generation introduces untrusted data into language models.
  • Malicious instructions in indexed sources can be executed with app authority.
  • Securing RAG pipeline requires access control, tainted data treatment, and output guarding.

More from Tuesday 1 September →