wlg1

View My GitHub Profile

Banner

Educational Outreach Overview and Plan

As a TA and lecturer, I practiced identifying which aspects of communication worked best to convey information to students. I identified 3 main principles to follow when translating difficult concepts into more understandable terms for a layman audience:

  1. Before delving into details, the purpose of each concept is stated, relating it to previous concepts. For instance, in my video "THE AI AMONG US in Your Non-Euclidean Mind" that breaks down a paper regarding style editing in diffusion models, the main aim is to remove features such as "hair color" in AI-generated outputs by moving away from them in feature space, where features are geometically represented as directions. Then, instead of just listing the numerous properties of a tangent space, I relate it to this purpose in the big picture.
  2. I explain only the parts that are relevant to the main purpose at hand. For instance, though the concept of the pullback is very intricate, I distill it to its purpose of transporting tangent vectors between tangent spaces. This simplification is then explained more rigorously in later videos.
  3. I drill down on explaining unfamiliar terms, based on prerequisite assumptions for the average target audience member. For example, instead of just assuming the audience knows what a tangent vector is- which in the original paper, would otherwise go unexplained- I explain the properties of it that are directly relevant to the main topic currently being discussed.

Overall, I plan to create two types of videos:

  1. Straightforward, purely educational videos that clearly explain the purpose of each concept
  2. Sci-fi story-oriented videos about AI safety, contrasting beneficial AI and harmful AI, and how they use actual AI mathematics to understand and cure harmful disorders. These would employ AI voices to narrate concepts, and AI generated media to animate them.

As I have only made a few videos so far, I am still learning how to best utilize the youtube algorithm. Essentially, as uploading more videos increases the chances of views, I plan to upload more videos, with the next one being a submission to the 3Blue1Brown competition in the middle of August. I also plan to go over research papers and their code, interacting with on-screen avatars of AI assistants and human guests to break down these works from different perspectives.

This link showcases a selection of these videos.