[Epistemic Status: This is an artifact of my self study. I am using it to remember links and help manage my focus. As such, I don't expect anyone to fully read it. If you have particular interest or expertise, skip to the relevant sections, and please leave a comment, even just to say "good work/good luck". I'm hoping for a feeling of accountability and would like input from peers and mentors. This may also help to serve as a guide for others who wish to study in a similar way to me. ]

Previous Entry: SSJ #3

Review of 3rd Sprint

My goals for this sprint were:

• SSJ--1 -- Write
  • AIA Terminology Lit Review
  • Math in my AI Alignment Goals
• SSJ--2 -- Read
  • Shallow Review of Technical AI Safety 2024
• SSJ--3 -- Math
  • Do some Linear Algebra reading and practice.
• SSJ--4 -- Experimentation
  • Go through Transformers From Scratch.
• SSJ--5 -- Tooling
  • Do an informal literature review on MI Tooling and Data Visualization for High Dimensional Data.
  • Places to start for MI Tooling:
    • The Interpretability Toolkit
    • TransformerLens & Callum McDougall's guide for it.
    • Nostalgebraist’s transformer-utils library
    • Google PAIR’s Learning Interpretability Tool (LIT)
    • Google PAIR’s What-If Tool
    • Jesse Vig’s BERTViz
    • LOOM
    • CircuitVis
• SSJ--6 -- Social
  • Email math profs after finished writing "Math in my AI Alignment Goals".
  • Consider other places to find potential mentors.
    • Consider what kinds of feedback I am looking for.
    • Should I reach out specific people for general advice on my SSJ or only once I have specific questions for them and their work?
    • Make some posts in various forums asking for people willing to review and comment on my SSJ.
  • I am also networking with the goal of finding future funding or paid roles. Consider strategies for that.

So how did I do?

Daily Worklog

Sprint Summary

Overview

During the past months I neglected focus on any of my chosen goals in favour of ad-hoc applying to fellowships. This both that my goals are not in alignment with what I think is really important, but also that I have difficulty prioritizing overhead organization and documentation, which I think are valuable, so I would like to improve and get back to doing this journal work.

SSJ--1 -- Write

I still like the idea of doing an AIA Terminology Review. I am less convinced by the value of discussing Math in my AIA goals specifically, rather, multiple people have suggested I should focus on writing and communicating about my "Theory of Change" both focusing on the importance of my (potential) role and the importance of my agendas. I would also like to try putting together better article to briefly explain my OIS concept based on various conversations and thoughts, and finally write the MAAT article I was planning to write.

SSJ--2 -- Read

I started skimming this but didn't really engage and then got busy. I think it may be a good resource to be included in my AIA Terminology Review.

SSJ--3 -- Math

I didn't actually practice any linear algebra. It's much harder to do so when I'm not handing it in to a professor for marks. I did however pick up my copy of C.Kosniowski's "algebraic topology" which I find engaging. The concepts may or may not have value for my thinking on semantic spaces. I think I still have a shallow understanding of topology which I would like to deepen towards a solid understanding of manifolds.

SSJ--4 -- Go through Transformers From Scratch.

This still seems high value. I still haven't started.

SSJ--5 -- Literature review on MI Tooling and Etc...

I didn't focus on this. I'm torn between wanting to jump in and begin implementation work, however, a review beforehand is probably quite valuable.

SSJ--6 -- Social

The focuses in this section may have been of higher benefit than applying to the various fellowships I have applied to. Additionally, I often got an email about a fellowship deadline, worked on an application before the deadline and then submitted it and moved on to something else. This approach seems somewhat aimless and disorganized. I would like to be keeping a better track of which fellowships I have targeted with what levels of effort and also have a better sense of what fellowships are out there and what other options I can consider.

Goals for 4th Sprint

The Goals:

• SSJ--1 -- Write
  • Make an article or doc to contain and organize articles I would like to write.
  • Theory of Change
  • OIS explainer
  • MAAT
  • AIA Terminology Review
• SSJ--2 -- Read
  • Search and read various articles for AIA Terminology Review.
  • Spend some time reading and comment on one random LW article 4 days / week.
• SSJ--3 -- Math
  • Low priority: Continue reading C.Kosniowski's "algebraic topology"
• SSJ--4 -- Experimentation (copied from last sprint)
  • Go through Transformers From Scratch.
    • SERIOUSLY! Clock in some time on this!
• SSJ--5 -- Tooling (copied from last sprint)
  • Do an informal literature review on MI Tooling and Data Visualization for High Dimensional Data.
  • Places to start for MI Tooling:
    • The Interpretability Toolkit
    • TransformerLens & Callum McDougall's guide for it.
    • Nostalgebraist’s transformer-utils library
    • Google PAIR’s Learning Interpretability Tool (LIT)
    • Google PAIR’s What-If Tool
    • Jesse Vig’s BERTViz
    • LOOM
    • CircuitVis
• SSJ--6 -- Social
  • Develop my networking plan
    • Create a list of people I respect who may be worth reaching out to for mentorship or networking.
    • Research and reach out to people where possible and pragmatic.
    • Clarify the problems I am interested in focusing on and the capacity in which I am interested in focusing on them. (High overlap with SSJ--1 "Theory of Change" )

Well. The last sprint went of the rails. Somewhat disappointing but I am wishing myself luck getting back on track.

List of common acronyms:

• Mechanistic Interpretability (MI)
• AI Alignment (AIA)
• Outcome Influencing System (OIS)
• n-Dimensional Interactive Scatter Plot (NDISP)
• Machine Learning (ML)
• Large Language Model (LLM)