A Reasonable Approximation

Alice: Are there any examples of X?

Co-authored with Linda Linsefors If you're training any type of toy model of superposition, Mean Squared Error (MSE) loss is unusually bad.1 Related work We aren't the first to notice that MSE loss doesn't work.

In September 2024, I started working on a patch to GHC.

About a month ago I left my job.

On Saturday (Feb 28, 2026) I attended my first ever protest.

I'm a big fan of the game Zendo.

I've previously complained about how people often repeat a quote that starts with The reason that the rich were so rich, Vimes reasoned, was because they managed to spend less money.

I previously wrote about Boots theory, the idea that "the rich are so rich because they spend less money".

I sometimes see people describe the Elm community as "very active".

Translated by Emily Wilson 1.

Quick note about a thing I didn't properly realize until recently.

Mostly out of curiosity, I've been looking into how cryptocurrency is taxed in the UK.

Some months ago I got a 3D printer.

Suppose you have a game where you can bet any amount of money.

I gave a fifteen minute talk about this at Zurihac 2023.

According to Wikipedia, the Shard (the tallest building in the UK) stands "309.6 meters (1,016 feet) high".

I've heard that descaling a kettle makes it more efficient, and can save me time and money.

t ≈ 0s I am born into knowledge.

One-sentence summary: Kelly is not about optimizing a utility function; in general I recommend you either stop pretending you have one of those, or stop talking about Kelly.

I don't know how many people read this blog1, but it doesn't receive many comments.