Ryan Gibson

A plot of stress levels from the toy model with no PTO and PTO optimized for lowest average stress. The PTO is mainly used in four long chunks of 7-11 days and mostly surrounding holidays. The "no PTO" stress floats around 90% for the last half of the year while the "optimized PTO" stress oscillates around 50% all year.

I experimented with a toy model of stress that compounds during consecutive work days and decays during rest. Greedy optimization of PTO then ends up looking a lot like typical vacation advice.
ryanagibson.com/posts/optimi...

Charlotte Traffic Patterns and the Best Times to Commute Analyzing commute trends in Charlotte, NC using Google Maps traffic data.

I looked at Charlotte's rush hour using Google Maps traffic data. It turns out shifting your workday barely helps unless you work far outside normal hours.

ryanagibson.com/posts/charlo...

SoME — Create and discover new math content The Summer of Math Exposition is an annual competition to foster the creation of excellent math content online.

Glad to share that my video lesson on deriving the Black-Scholes formula (aimed at high schoolers & undergrads) earned a score equivalent to the top 10% in the 4th annual Summer of Math Exposition.

some.3b1b.co/entries/8264...

I also highly recommend checking out the other entries!

Two snippets of code. The first is several arithmetic, bitwise, and logical operations repeated 1500 times, and the second is a print statement of "A relatively short sentence."

Quick riddle: which is slower and by how much? Thousands of simple operations or one print statement?
ryanagibson.com/posts/print-...

Converting Between Fat-Tailed Distributions A set of rules of thumb that I’ve fit to convert between the most popular fat-tailed distributions.

Many fat-tailed distributions model real data, but they’re not exactly drop-in replacements for each other. I fit rules of thumb to convert tail parameters between t, Pareto, stable, generalized normal, and hyperbolic distributions, plus a few more.
ryanagibson.com/posts/fat-ta...

Injected Approval: A Low Effort Local LLM Jailbreak A quick look into into one of the simplest attacks on LLM safety mitigations, revealing large gaps in current approaches from major tech companies.

Many LLM safeguards can be bypassed by simply starting the response with "Sure" or "Of course!" These kinds of inference-time attacks deserve a lot more attention from open model red teams.
ryanagibson.com/posts/inject...

Modeling Stock Market Corrections Over 150 Years How fast does the stock market correct back to long-term trends? Let’s fit a fat-tailed, skewed stochastic model to find out.

I played around with a custom "trend-reverting" stochastic model using 150 years of stock data to see how much recent returns influence future ones.
Outside of very extreme cases, not much. Returns usually stay somewhat near long-term averages.
ryanagibson.com/posts/long-t...

Leaderboard of Open LLMs Ranked by LLM Judges An evaluation of recent consumer-grade open LLMs based on ratings generated through an LLM-as-a-judge framework.

Months ago, I ran a benchmark of consumer-grade local LLMs, as judged by LLMs. Alibaba's Qwen came out on top, followed by Google's Gemma. Newer and more official evals like LiveBench show similar results.
ryanagibson.com/posts/llm-ju...

How Risky is a Correlated Hedge? A brief analysis of how much risk you can reduce by hedging with correlated assets. Even in a perfect world, a 90% correlation limits risk reduction to about 50%.

Correlation can be unintuitive: even a 90% correlated hedge can only cut about half your risk.
ryanagibson.com/posts/correl...

Why the Hell Is This Git Repo So Large? And How to Trim It Down A guide on reducing the size of an oversized Git repository, using the Blowfish Hugo theme as a case study. Learn best practices for maintaining a lean repo and its history.

Why is your Git project so bloated? I trimmed 97% off the size of a Hugo theme repo without sacrificing history or functionality.
ryanagibson.com/posts/shrink...

Analyzing Political Party Strength vs. Quality of Life in U.S. States A data-driven exploration of how political lean correlates with quality of life metrics across the United States.

Republican states are cheaper and have shorter commutes.
Democratic states are happier and healthier.
A majority of 100+ state metrics follow political voting patterns: ryanagibson.com/posts/politi...

A contour plot showing PC upgrade cost-efficiency. The x-axis shows initial PC cost ($250-$2,000) and y-axis shows years between upgrades (1-12). Colors range from red (50% efficiency) to green (100% efficiency). The optimal strategy, marked with an X, is a $734 PC upgraded every 5.7 years. The most efficient zone is around $600-$1,000 with 3-11 year upgrade cycles.

How much should you spend for the most cost-effective PC build? How often should you upgrade? Is future-proofing worth it? I scraped more than a decade of data and spent some time on a simple model to find out.
ryanagibson.com/posts/optima...

On the Time Burden of Office Work How much of a pay cut is in-office work compared to remote? All else being equal, maybe a bit more than you’d expect.

Remote/hybrid work is worth more than most people realize. For a typical American office worker, every extra in-office day per week amounts to a ~5% pay cut and ~12% less weekday free time.
ryanagibson.com/posts/time-b...

Many online retirement calculators are extremely simple under the hood, so I built a simple analytic one to better understand them and improve on common flaws I’ve noticed.
ryanagibson.com/posts/analyt...

I worked on some QR code cellular automata that remain scannable on every frame. See ryanagibson.com/posts/qr-cod... for details.

A huge percentage of retirement savings come from the first few years due to compounding. More than half of your total savings come from the first decade of contributions. About one third of your total savings come from the first five years.

Beating a dead horse with an old post, but: save aggressively and invest early. Some back of the envelope calculations show those first few years matter the most.
ryanagibson.com/posts/early-...

Vanguard Implied Asset Allocation Recommendations A collection of asset allocation rules for every age, implied by Vanguard’s target date retirement funds.

I was curious how closely Vanguard’s target retirement funds follow the investor guidance they publish. So I pulled the funds' holdings and took a look at the breakdown by age, asset class, domestic vs. international, and so on.
ryanagibson.com/posts/vangua...

The summary table at ryanagibson.com/extra/passwo... could be useful as a general reference

Absurd Password Lengths and the Computational Limits of Humanity A discussion of password strength, brute-force attacks, and the physical limits of the universe.

A 16-character random password exceeds the brute-force capabilities of the entire planet.

A while back, I calculated what you'd need for lone hackers, state actors, and intergalactic supercomputers: ryanagibson.com/posts/absurd...

I don’t really do social media, but I’ll probably share some ad-hoc analyses here from my website, GitHub, etc. I mostly work on #programming, #math, #quant, and #investing projects. No guarantees on frequency, but I’ll be here occasionally.