Applied Mathematician, probabilist, software engineer.
Bay Area
Joined July 2011
- Tweets176
- Following304
- Followers110
- Likes1.3K
Ian Langmore retweeted
This is a dream role for any lawyer interested in tech and free speech, and who wants to work across litigation, legislative & policy, and communications.
.@TheFIREorg is hiring!
Lead Counsel, Tech and Free Expression Policy
This person will lead FIRE's strategy for ensuring free speech protections apply to emerging communications technologies.
Ian Langmore retweeted
Last night, FIRE filed a friend-of-the-court brief with the U.S. District Court of Northern California arguing the Department of War's retaliatory designation of Anthropic as a supply chain risk violates the First Amendment.
@EFF, @CatoInstitute, @ProgressChamber, and the First Amendment Lawyers Association joined FIRE on the filing.
Ian Langmore retweeted
We irritate a lot of partisans by insisting on the consistent protection of free speech.
Ian Langmore retweeted
The @nytimes been covering this since 2022.
nytimes.com/2025/11/29/us/fr…
nytimes.com/2024/06/07/us/fe…
nytimes.com/2024/06/03/us/ju…
nytimes.com/2025/12/18/us/mi…
nytimes.com/2022/03/08/us/po…
Ian Langmore retweeted
I first read about it in The New York Times a month ago.
nytimes.com/2025/11/29/us/fr…
You literally can’t find a single mainstream article about the Minnesota fraud...
Not even one...
How is it that a simple guy with a camera can uncover over $110,000,000 in fraud in a single day, but giant billion‑dollar media organizations behind the glass doors somehow “miss” it for years?
It’s not that they couldn’t find it. It’s that they wouldn’t
They are desperate to bury this, protect their ideological allies, and smear anyone who exposes how rotten the system is with “it’s not as bad as it looks” think‑pieces or by turning it into a race‑based narrative to shut people up
This is as clear as it gets in broad daylight
The story isn’t just the fraud...
The story is the massive cover‑up....
Ian Langmore retweeted
Moment Expansions of the Energy Distance. arxiv.org/abs/2505.20647
Ian Langmore retweeted
Ian Langmore: Random Field Representations of Kernel Distances arxiv.org/abs/2505.23141 arxiv.org/pdf/2505.23141 arxiv.org/html/2505.23141
This week I posted two papers to arXiv. They were motivated by the question, “what does the energy distance do in high dimensions?” The first is a pretty straightforward attempt: Moment Expansions of the Energy Distance. arxiv.org/abs/2505.20647
Working on this second paper led me to some of the most interesting math I’ve ever done. Thanks to Srinivas Vasudevan for leading the stochastic processes reading group, and former teammates @dkochkov1, @janniyuval, @shoyer
...for the weather forecast work (NeuralGCM, nature.com/articles/s41586-0……) that inspired the original question.
Ian Langmore retweeted
Mathematics is not a careful march down a well-cleared highway, but a journey into a strange wilderness, where the explorers often get lost. Rigour should be a signal to the historian that the maps have been made, and the real explorers have gone elsewhere.
- W. S. Anglin, as mentioned in Mathematical Intelligencer, 1992
Ian Langmore retweeted
New @nature paper: nature.com/articles/s41586-0…
NeuralGCM results (all are a "first"):
1) A differentiable hybrid atmospheric model
2) Competitive with ECMWF ensemble
3) Competitive with a GCRM in a year-long simulation
4) 40-year AMIP-like runs. Smaller bias than AMIP runs.
1/
Ian Langmore retweeted
I'm incredibly proud to share NeuralGCM, our new AI and physics based approach to weather and climate modeling with state-of-the-art accuracy, published today in @Nature:
nature.com/articles/s41586-0…
Ian Langmore retweeted
I was surprised by a talk Yejin Choi (an NLP expert) gave yesterday in Berkeley, on some surprising weaknesses of GPT4:
As many humans know, 237*757=179,409
but GPT4 said 179,289.
For the easy problem of multiplying two 3 digit numbers, they measured GPT4 accuracy being only 59% accuracy on 3 digit number multiplication. Only 4% on 4 digit number multiplication and zero on 5x5. Adding scratchpad helped GPT4 but only to 92% accuracy on multiplying two 3 digit numbers.
Even more surprisingly, finetuning GPT3 on 1.8m examples of 3 digit multiplication still only gives 55 percent test accuracy (in distribution).
¯\_(⊙︿⊙)_/¯
So whats going on? Multiplication is algorithmically very challenging (as are less known algorithmic problems).
The authors hypothesize that Transformers have a hard time because they learn linear patterns that they can memorize, maybe compose, but not generally reason with. The paper raises interesting theoretical and practical questions on understanding what Transformers can learn.
The paper
"Faith and Fate: Limits of Transformers on Compositionality" says:
"Our empirical findings suggest that Transformers
solve compositional tasks by reducing multi-step compositional reasoning into
linearized subgraph matching, without necessarily developing systematic problem solving skills"
Ian Langmore retweeted
I suspect GPT-4's performance is influenced by data contamination, at least on Codeforces.
Of the easiest problems on Codeforces, it solved 10/10 pre-2021 problems and 0/10 recent problems.
This strongly points to contamination.
1/4
