Mikael Brunila
mikaelbrunila.bsky.social
Mikael Brunila
@mikaelbrunila.bsky.social
Postdoctoral researcher, author, and programmer. Most of my work is in #NLProc, #GIS, computational social science, and philosophy of technology.

https://mikaelbrunila.com/information-theory/
Pinned
Chapter three from my thesis is finally out in @bigdatasoc.bsky.social! I explore the parallels between the "bit" and the "embeddings" that are a foundational structure in Large Language Models, showing how the latter endow tech companies with what I call "cosine capital."

bsky.app/profile/bigd...
🚨 New in Big Data & Society 🚨
"Cosine capital: Large language models and the embedding of all things" by Mikael Brunila doi.org/10.1177/2053...

Proposes “cosine capital” to describe how LLM embeddings commodify language and data, reshaping power, abstraction, and AI economies.
"Devs" just might be the best example of Laplace's demon in pop culture
December 4, 2025 at 3:49 AM
1999 really was the year of the anti-work movie. Office Space, The Matrix, Fight Club... anything else?
December 3, 2025 at 5:03 PM
it's funny that GPT models are called "decoder-only" when almost all of the computation is actually about encoding the input sequence ("the cat sat on the") into a contextual representation–and only the very last matrix maps that representation into next-token predictions ("mat").
December 1, 2025 at 5:44 PM
Reposted by Mikael Brunila
gather around, kids, and i'm going to tell you about a time when people used to develop models that they understood the inner workings of, and they would use those models to test and evaluate their understanding of complex systems...
a cartoon of homer simpson with the words which was the style at the time
Alt: abe simpson in union civil war attire with some of his friends saying "which was the style at the time."
media.tenor.com
November 30, 2025 at 6:24 PM
Reposted by Mikael Brunila
We recently released OlmoEarth—an open spatio-temporal foundation model for planetary intelligence, trained on multimodal Earth observation, OpenStreetMap, & other open data. It sets a new accuracy/efficiency Pareto frontier and powers our OlmoEarth Platform. Technical report now on arXiv. 👇
November 26, 2025 at 7:23 PM
is there anything interesting written on how LLM alignment techniques implement microeconomic utility theory?
November 25, 2025 at 9:08 PM
Anna's Archive is great, but why do half of the downloads consistently fail?
November 25, 2025 at 5:48 PM
the Notes app autocorrects "nomos" into "gnomes"...

on the other hand, I guess both concepts are kind of "earthy"?
November 23, 2025 at 6:52 PM
I wrote a comprehensive introduction to basic information theory, with added context from social scientists of the cybernetics movement (Gregory Bateson) & later (Katherine Hayles). It covers everything from bit encoding to comparative metrics like cross-entropy.

mikaelbrunila.com/information-...
From Bits to Embeddings – A Critical Introduction to Information Theory
A comprehensive introduction to classical information theory as conceived by Claude Shannon and its connections to neural networks and large language models today.
mikaelbrunila.com
November 20, 2025 at 5:24 PM
"world model"
November 13, 2025 at 1:45 AM
Chapter three from my thesis is finally out in @bigdatasoc.bsky.social! I explore the parallels between the "bit" and the "embeddings" that are a foundational structure in Large Language Models, showing how the latter endow tech companies with what I call "cosine capital."

bsky.app/profile/bigd...
🚨 New in Big Data & Society 🚨
"Cosine capital: Large language models and the embedding of all things" by Mikael Brunila doi.org/10.1177/2053...

Proposes “cosine capital” to describe how LLM embeddings commodify language and data, reshaping power, abstraction, and AI economies.
October 22, 2025 at 4:28 PM
Cursor is great for greenfield coding a new project, especially if you're not very fluent in that language, but the autocomplete is surprisingly bad
October 20, 2025 at 3:50 PM
Reposted by Mikael Brunila
Why is this horrible for Universities?

Just about everyone who isn't a citizen or green card holder already who's hired for a tenure track faculty position is hired through an H1B and then, after 3-5 years, applies for a green card.

This is literally "No more foreign professors can be hired"
Those on an H1B cannot return to the US from tomorrow (Sunday) unless paying $100K. This is an out-of-the blue presidential action. We’ll see software engineers stranded abroad.

One easy to predict outcome: those on US visas will travel less… for work, for conferences etc.
September 20, 2025 at 2:22 PM
Reposted by Mikael Brunila
In a stunning moment of self-delusion, the Wall Street Journal headline writers admitted that they don't know how LLM chatbots work.
July 21, 2025 at 1:48 AM
Reposted by Mikael Brunila
IDF soldiers have now themselves admitted they have been ordered to murder aid-seekers in Gaza.

Truly an atrocity built on atrocity.

www.haaretz.com/israel-news/...
June 27, 2025 at 12:05 PM
it seems like almost every Silicon Valley adjacent book needs to mention Dunbar’s number. Alex Karp’s (Palantir) ”The Technological Republic” was no exception
May 5, 2025 at 3:09 AM
the problem with most continental theory today is that it’s a lot of ”tell” and almost no ”show”. Incredibly tedious to read.
May 3, 2025 at 6:41 PM
Reposted by Mikael Brunila
existence of tensors implies existence of relaxors
April 22, 2025 at 1:38 PM
Reposted by Mikael Brunila
The key chart right now:

Usually US economic pain is cushioned by falling bond yields and a strengthening dollar, which mean lower interest rates and more spending power for consumers.

This time we’re seeing the opposite, meaning the pain will be amplified.
April 22, 2025 at 1:41 PM
Reposted by Mikael Brunila
Evidence of US brain drain?

US scientists submitted 32% more applications for jobs abroad Jan-Mar 2025 than during same period in 2024.

US-based users browsing jobs abroad increased by 35%.

Data from Nature Careers global science jobs platform.
www.nature.com/articles/d41...
April 22, 2025 at 10:05 AM
Reposted by Mikael Brunila
With all respect to the sanguine predictions from fellow economists: Standard quantitative trade models don’t seem to capture the loss of global public goods, such as the rules-based international order.
April 7, 2025 at 12:15 AM
Reposted by Mikael Brunila
I just don't think there's evidence that the direction of LLM ability-increase is towards human intelligence or a superset of human intelligence. Really seems like they are their own thing that partially overlaps with human intelligence
March 10, 2025 at 2:10 PM
Reposted by Mikael Brunila
March 8, 2025 at 3:51 PM