Meta-Notes

This is a republish of a previous post, after the previous version went through heavy editing, updates and changes. The text has been expanded, content moved around/added/deleted.

Estimated reading time: 2 Hours 40 minutes reading at 100 wpm. Given the density of material covered in this chapter, if someone is encountering these arguments for the first time, I think the lesswrong projected time is too low.

Estimated reading time (including Appendices): 3 Hours 40 minutes reading at 100 wpm. The appendices are meant for people already quite familiar with topics in machine learning.

If you feel there are any mistakes or misrepresentations of anyone's views in any section please let us know.

0.0: Overview

State-of-the-Art AI. We begin with a brief introduction to the current advancements in artificial intelligence as of 2024. We aim to acquaint readers with the latest breakthroughs across various domains such as language processing, vision, and robotics.

Foundation Models. The second section focuses on foundation models, the paradigm powering the state-of-the-art systems introduced in the previous section. We explain the key techniques underpinning the huge success of these models such as self-supervised learning, zero-shot learning, and fine-tuning. The section concludes by looking at the risks that the foundation model paradigm could pose such as power centralization, homogenization, and the potential for emergent capabilities.

Terminology. Before diving deeper, we establish the definitions that this book will be working with. This section explains why "capabilities" rather than "intelligence" is a more pragmatic measure for discussing AI risks. We also delineate key terms within the AI debate, such as Artificial General Intelligence (AGI), Artificial Super Intelligence (ASI), and Transformative AI (TAI). The section concludes by introducing the (t,n)-AGI framework which allows us to more concretely measure the level of AI capabilities on a continuous scale, rather than having to rely on discrete thresholds.

Leveraging Computation. In this section, we explore the importance of computation in AI's progress introducing the three main variables that govern the capabilities of today's foundation models - compute, data and parameter count. We explore scaling laws and hypotheses that predict the future capabilities of AI based on current scaling trend...

Denis Shmurak

Denis Shmurak