[Epistemic Status: This is an artifact of my self study. I am using 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 #4

Highlights

I published a post to make my planned writing publicly visible.
I wrote and published Agent Foundations: Paradigmatizing in Math and Science.
I started writing a sequence: Map articulating all talking (Maat).
I attended my BSc Convocation Ceremony! 🎉 Computer Science Honours with Math Minor : )

Review of 4th Sprint

My goals for this sprint were:

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" )

So how did I do?

Daily Worklog

Date	Progress
Th, Nov 6	Finished and published SSJ #4 Set up SSJ #5 doc Read and commented on Legible vs. Illegible AI Safety Problems Skimmed some popular AI books while at the library. May post some thoughts on them sometime. Started concept mapping for my "Theory of Change".
Fr, Nov 7	Read about some Theory of Change concepts and terminology. Finished writing and posted "Agent Foundatiosprintns: Paradigmatizing in Math and Science"
Mo, Nov 10	Read and commented on Ontology for AI Cults and Cyborg Egregores. Transformers from Scratch (TfS): Went through and made sense of the structure and read through the overviews of each section. I will at least go through 1.1 and 1.2 in order, and then very likely continue through 1.3.1. I'd like to continue through the rest of the sections in order, but I'll judge based on how long it's taking me. I may skim the rest and only do the exercises later. Started Map articulating all talking (Maat) sequence. Started looking at high dimensional books and papers for SSJ-5
Tu, Nov 11	No progress because of commuting and errands.
Wd, Nov 12	No progress because of: 🎉 My BSc Convocation Ceremony! Computer Science Honours with Math Minor : )
Th, Nov 13 - Fr, Nov 14	Pretty burned out from convocation related socializing, so not much progress. Did some worth ideating and reading and commenting on some LW posts.
Mo, Nov 17	Planning and started writing next post in Maat sequence.
Tu, Nov 18	No progress. Distracted by interpersonal "other life stuff".
Wd, Nov 19	Finished writing "What the term "Mass Communication" gestures at" and requested editing.
Th, Nov 20	Watched the first TfS lecture. Revised "What the term "Mass Communication" gestures at" and posted it.

Sprint Summary

Overview

I have my BSc now 🎉

I'm also feeling good about the amount of focus on this I've had in first two weeks of the sprint, but then got very distracted and depressed by other things in my life and didn't manage to record any progress for the time from then til now.

So I definitely still want to improve, but it feels like I'm moving in the right direction. I also want to get better at prioritizing things that are worth working on, and sticking to my plan. This also means putting more realistic amounts of work in each sprint. I got similar advice from @Roman Malov and from reading A Pragmatic Vision for Interpretability, so I'm hyped up on that.

More object level, I have started the Transformers from Scratch (TfS) series, written a few posts, and read and commented on several posts.

SSJ--1 -- Write

I had five items on my write list. I'm realizing writing about my theory of change is probably more involved than I thought, so I ended up putting more focus into Maat, which I roughly know everything I want to write, I just need to write it down.

Also, I published a post to make my planned writing publicly visible. I may change the format later, but I like the idea of the list being public, so I'm committing to that.

SSJ--2 -- Read

I didn't do any of the review for AIA terminology review. I still think that's a good idea but is quite involved, so I probably should make it a main focus if I'm going to do it.

I did manage to read and comment on LW posts. I think this is good practice, so I plan to continue. I think looking at the list of all posts is a good way to engage with what other people are currently focusing on, however, I think there is value in reading and commenting on older posts and I'm not sure how to prioritize that very well. So one goal could be to figure out how to prioritize things to read, but that sounds dangerously meta.

SSJ--3 -- Math

I think I might have picked the book up once. I'm not having much traction staying engaged with math study now that I'm not in classes. It is such a difficult thing to do it might be better to treat it like the AIA terminology review, both in having a description of my understanding and some worked examples as a visible output, but also in that I should not try to do it unless it is a main focus.

SSJ--4 -- Experimentation

I actually started going through Transformers from Scratch! I haven't made it very far but I've finally actually started so that's nice.

SSJ--5 -- Tooling

I started looking at literature focused on high dimensional spaces. There's so much content out there it's difficult to get a sense of it all. One noteworthy resource I found is "Understanding High-Dimensional Spaces" by David B. Skillicorn. I definitely want to at least skim this.

SSJ--6 -- Social

I've been engaging with people here on LW which feels like movement in the right direction, but I definitely want to be more conscientious in my planning and execution in this domain. Clarifying my goals here would probably be good.

Goals for 5th Sprint

I think trying to focus on all 6 of my pursuit categories at the same time has made me too scattered. So going forward I will instead keep them as guides, but pick fewer specific goals from within each for each sprint.

The Goals:

Every day spend some time on each of the following:
- Read some LW post or other relevant material. (SSJ--2)
- Spend some time writing or developing ideas to write (SSJ--1)
- Work on Transformers from Scratch course (SSJ--2&4)
By the end of the sprint:
- Have clarified my SSJ--6, social networking, goals and strategy, and write a post describing them.

I'm hoping having fewer goals will make them easier to focus on. The "every day" goals make for an easy quantification: Either I worked on it that day or I didn't. I would like to be using better metrics, but I don't want to make up numbers that may not actually indicate anything useful. Numbers are good if they mean something, but if they don't, it's better to speak qualitatively.

List of common acronyms:

Mechanistic Interpretability (MI)
AI Alignment (AIA)
Outcome Influencing System (OIS)
n-Dimensional Interactive Scatter Plot (ndisp)
Map articulating all talking (Maat)
Machine Learning (ML)
Large Language Model (LLM)