Christopher Wolfram

Asking whether language models have human mental states is the wrong question.

LLMs are useful and interesting precisely because they can perform tasks for which robust algorithms are elusive.

LLMs make consistent mistakes when given prompts that are rephrased or in another language.

Independently-trained large language models generate activations that are similar between models, but not within models.

When an LLM is trained to be good at one task, will it also get better at other tasks?

I test a claim that the Ugaritic alphabet (one of the oldest alphabets) was designed to use fewer strokes to write the most frequently used letters.

In the next few years, the most promising role for LLMs in academia will be applications, along with some research of how LLMs work.

Claudius Ptolemy was one of the most influential scientists of all time, but more recent analysis of one of his most important works led to claims of plagiarism.

Politics can often be reduced to a single Left-Right dimension.

By giving LLMs the same resources as human programmers, they can use more sophisticated, iterative programming workflows, and write more reliable code.

The goal of this project is to use survival analysis to study the lifespan of information on Wikipedia and Wikidata with the goal of understanding 1) how lasting contributions are made in collaborative projects, and 2) the ‘half-life’ of knowledge.

Characterizing ecosystems looking at the set of species that inhabit them by using 500GB of citizen science data.

By applying hierarchical community finding algorithms from network analysis to commuting flows from the US Census, it is possible to characterize and map socially-meaningful communities from giant megaregions, down to neighborhoods.

During the course of another project, I wrote Rust code for parsing the entire ~500GB bitcoin blockchain and for extracting data on mining centralization.

The goal of this project is to apply methods developed in a previous project to characterize the geopolitical community structure of the UN General Assembly.

The eight queens puzzle is a classic chess problem with the goal of placing eight queens on a chess board so that no two can attack each other.

Over the last 230 years or so, the US Congress has met 117 times.

Data from NASA satellites shows how Ukraine has gotten darker in the weeks following the Russian invasion.

The Spelling Bee is a word puzzle published daily by The New York Times.

For any two points on the Earth, the geodesic (as the crow flies) distance between them is usually different from the travel distance (along roads).

Voronoi diagrams are a useful geometric tool across a number of applications.

For any two points on the Earth, the geodesic (as the crow flies) distance between them is usually different from the travel distance (along roads).

In March of 2020, at the beginning of the COVID-19 pandemic in the US, there was endless talk of disease spread modeling.

I couldn’t find a high-resolution map of household income for the whole United States, so I decided to make one.

Rush Hour is a logic puzzle board game in which the player slides “cars” around a grid with the goal of driving a particular target car to the exit.

News stories have a life cycle: they break, they are covered for a while, and then they fade into irrelevance.

Where is the geographic center of the US?

I trained a variational autoencoder (VAE) on a set of thousands of flags from around the world, making it possible to generate new flags and to measure the “probability” of existing ones.

Some logical connectives (boolean functions like AND, OR, and NAND) are functionally complete, which means they can be used to emulate any of the other logical connectives.

Since 1958, Billboard magazine has been publishing its Hot 100 list of the 100 “hottest” songs every week (based on sales, radio play, streaming, etc.)

In his book Guns, Germs, and Steel, Jared Diamond proposed that societies grow in accordance with regional differences in climate, and that because climate varies more drastically when moving north-south than when moving east-west, large societies will gr…

In 2015, CNN-based OCR was in its infancy, and traditional tools for OCR relied on there being a single baseline for the target text.

I worked as a science consultant on the science fiction movie Arrival (2016) which centers around a linguist trying to understand an alien language.

Back in 2015, while I was a Sophomore at the Stanford Online High School (OHS), I wrote a series of programs for extracting and analyzing data from recordings of my online classes.

Boids is a simple model of flocking behavior which has been used to simulate the movements of birds in a flock, fish in a school, etc.

I was the primary developer of the original cryptography functions shipped with every copy of Wolfram Language (Mathematica) since 2015 (version 10.1).

I created the original version of the Wolfram Tweet-a-Program (TAP) bot.

Back in 2013, I implemented several maze generation algorithms.