Join getAbstract to access the summary!

Large Language Models

Join getAbstract to access the summary!

Large Language Models

MIT Press,

15 min read
8 take-aways
Audio & text

What's inside?

Are you using Large Language Models (LLMs) but aren’t sure how they work? Learn the basics of this transformative tech.

  • 0.5×
  • 0.75×
  • 1×
  • 1.25×
  • 1.5×
  • 1.75×
1×

Editorial Rating

7

getAbstract Rating

  • Scientific
  • Applicable
  • Background

Recommendation

LLMs are growing in complexity and scale: They can already generate sophisticated and complex content and perform some jobs faster and more efficiently than humans. But while this transformative tech is becoming ever more ubiquitous, few people actually understand how it works. In this helpful guide, computer science professor and natural language processing expert Stephan Raaijmakers demystifies LLMs, ensuring readers gain a concrete understanding of these AI tools. Learn the basics of neural networks and how LLMs predict responses, as well as the potential risks of this powerful new technology.

Summary

Stop viewing LLMs as “black boxes.” Invest time learning how they work.

Over the past few years, large language models (LLMs) have become increasingly ubiquitous, yet they often function as “black boxes,” primarily because few people have a solid understanding of how they work, though also because systematic standards don’t yet exist for AI tools and software products.

Generative LLMs use “vector embeddings,” or lists of numbers, to represent words, and predict the next “token” (generally a word or part of a word). LLMs learn these vector embeddings using artificial neural networks, and they leverage an “attention mechanism” to determine the contextual relationships between each word. Together, the word embeddings, neural networks, and attention mechanism of each LLM comprise what’s commonly referred to as “transformer architecture.”

There are different types of LLMs that serve different purposes. For example, you can use generative models to ideate and produce new inputs, and you can use classifier models to predict labels — that is, the group that an object belongs to. Whatever the type, these models are only...

About the Author

Stephan Raaijmakers is a professor of Communicative AI at the Centre for Linguistics.

Comment on this summary