2025 Goals

My primary goal for 2025 is to improve my AI research engineering skills. While my expertise is in LLM evaluations, I recognize that evaluations alone are not sufficient to make AI safe. It seems like AGI could be here within approximately three years, so 2025 might be the last year I have to spend significant time on foundational upskilling. I will primarily focus on improving my math automaticity and completing the ARENA program.

I also have a few secondary goals to support that primary goal, such as staying organized and writing on this website. In addition, for work–life balance, I will exercise, meet up with friends, and meditate regularly. I also track a few behaviors I want to largely avoid.

To stay on track with these goals, I've captured them in Beeminder and published them here for accountability.

Improve my mathematical automaticity

My math skills are sufficient for any engineering work I need to do, but they aren't yet at the level where I feel confident doing deep, mathematically heavy AI safety research. In high school and university, I was good at learning for tests. As a result, I never deeply understood mathematics at the level I believe is warranted, and much of my skill has since atrophied.

To remedy this, I've been using Math Academy to sharpen my skills. I'm committing to 80XP per day, on average, for the next year. This should allow me to complete the Mathematics for Machine Learning course and ensure automaticity in all the foundational skills below that level.

Additionally, for more applied math, I plan to solve one Project Euler problem every three days or so, on average.

The chart on the left represents total XP on Math Academy, and the chart on the right is cumulative Project Euler problems solved.

Complete the ARENA 3.0 program

The ARENA (Alignment Research Engineer Accelerator) program is a fantastic resource. I started to do this in Februrary 2024 after Joseph Bloom encouraged me to, but I got distracted with other material and never finished it. A few of us at AISI are doing this to develop and refresh our research engineering skills, and I intend to complete it this time around.

The chart on the left represents commits to my ARENA repo, and the chart on the right represents days I made at least one commit.

Exercise every day

I've found that one of the most important factors for sleep, focus, and general cognitive abilities is exercise. For this, I am tracking two goals:

Work out at least 30 minutes a day, on average.
Work out every day for any amount of time (as measured by a meta-Beeminder goal based on the first goal).

The second goal ensures consistency, preventing me from "banking" a bunch of exercise in one day and then skipping the next. Consistency greatly helps with habit formation.

The chart on the left shows total workout minutes (starting in November 2024), and the chart on the right shows the number of days with workouts logged in 2025.

Stay Organized

Maintaining a high level of organization is essential to avoid being overwhelmed. I have three primary organizational practices:

Clear inboxes (Gmail, messages, etc.) and move any non-trivial tasks into Todoist.
Keep Todoist organized, with tasks categorized and scheduled.
Time-block plan each day to ensure effective execution of my tasks.

Each practice is not difficult by itself, but maintaining them consistently can be challenging. So, I'm tracking them below. Luckily, Todoist has a great API, which makes this straightforward.

The chart on the left shows how many tasks are currently in my Todoist inbox. The chart on the right shows how many tasks I complete over time. I use a simple Google Cloud Run Python script to calculate the number of inbox tasks, which Beeminder then fetches.

The chart on the left shows how many inboxes I've cleared daily. The chart on the right, inspired by Cal Newport, tracks how often I time-block plan my day. I have a recurring Todoist task which populates this data.

Update this website daily

I originally created this website in 2014. Working on it has always brought me joy, though I've let it lapse except for brief periods in 2016 and 2020. As I no longer use social media, I appear mostly absent online if I'm not publishing here.

I've come across several compelling arguments about why writing online is important—especially now that writers' content is increasingly used to train AI models (Gwern and Derek Sivers have both discussed this.) Because I already journal privately to organize my thoughts, I might as well share much of that writing publicly to help train future AI systems in line with my values.

The chart on the left displays commits to this website (starting in October 2024), and the one on the right is 2025 days with at least one commit.

Meet up with friends

I recently moved to London and haven't yet developed a broad social network here. With the demands of work and these goals, it can be challenging to make time for socializing, but it is important nonetheless. I specifically aim to make time for friends I don't see on a regular basis.

This chart shows meetings with friends I don't regularly see.

Meditate

Given the rapid pace of change in my life, I've found it helpful to take time just to focus on my breath.

This chart tracks minutes spent meditating.

Behaviors to minimize

Equally important to tracking behaviors I want to do more of is tracking the behaviors I want to minimize.

Concerning substances, I occasionally use alcohol and caffeine. Alcohol has never been a big part of my life, but British drinking culture can be quite intense and local pubs are quite cozy, so it's worth monitoring. Caffeine, however, has been more of an issue—I typically consume large amounts of tea daily, which affects my sleep. In 2025, I intend to wean myself off of caffeine. I still plan to enjoy a drink or coffee when needed, but only once a week or for special social occasions.

The chart on the left tracks alcoholic drinks consumed (2 per month limit). The chart on the right tracks caffeinated drinks consumed (1 per week limit).

Regarding technology usage, I'm usually good at avoiding video games and time-wasting websites (Reddit, HN, YouTube, etc.). However, I find them difficult to quit once I start again, so I'm making sure to keep these in check.

The chart on the left tracks video game hours (2 per month), and the chart on the right tracks visits to "junk" websites (2 per month).