Website Source: blog / arrakis_blog

Summary

Pending synthesis from local website source.

Original source title: Step1: Create a function where you can do operations on the model.

Extracted Preview

Note: Arrakis is a work which is really personal and pivotal for me. I worked on it solo during my free time while simultaneously working a co-op and uni. Buildspace was really pivotal in providing the push.

Arrakis - A toolkit to conduct, track and visualize mechanistic interpretability experiments.

Project Name : Arrakis.
Project Description : Arrakis is a library to conduct, track and visualize mechanistic interpretability experiments.
Project Home : [Github Repo](https://github.com/yash-srivastava19/arrakis)
Project Documentation : [Read the Docs](https://arrakis-mi.readthedocs.io/en/latest/README.html)
PyPI Home : [PyPi Home](https://pypi.org/project/arrakis-mi/)

Introduction

_"The greatest enemy of knowledge is not ignorance, it is the illusion of knowledge."_

-Daniel J. Boorstin

Understanding how we think is a question that has perplexed us for a long time. There have been countless theories, many thought experiments, and a lot of experiments to try to unravel the mysteries of the brain. With time, we have become more aware of our brain's working, and in my honest, and a little biased opinion, Artificial Intelligence has come the closest to model our mystery organ.

This is one of the reasons why Interpretability as a field makes a lot sense to me. It tries to unravel the inner working of one of the most successful proxy of human brain - Large Language Model. Mechanistic Interpretability is on the approach in AI alignment to reverse engineer neural networks and understand the inner workings.

Integration Notes

Source section: blog
Local source: /home/yashs/Desktop/Programming/yash_blog/yash-srivastava19.github.io/blog/arrakis_blog.md
Raw copy: raw/website/yash-srivastava19-github-io/blog/arrakis_blog.md