-
Book Overview & Buying
-
Table Of Contents
-
Feedback & Rating
Threat Modeling Best Practices
By :
In this chapter, given the fluidity of the field, we will focus mainly on language models. The reason is that most of the AI we interact with today has been trained on and utilize one or many models to produce its results. They serve as the language processing component of broader AI ecosystems, contributing to tasks such as text generation, translation, question-answering, and reasoning. We’ll now look into a few types of language models and their unique risks, which could help us develop threat models to protect them.
These models follow the original mathematical approach to language processing, based on calculating probabilities of word sequences from a text corpus. These models rely on frequency counts and assumptions to predict the likelihood of words appearing in specific contexts, forming the foundation of early natural language processing (NLP) before the deep learning revolution. A statistical language model...
Change the font size
Change margin width
Change background colour