It is so painfully obvious that a more sensible world would be treating the development of AI with much more care than what we see from the current state of affairs.
I really wish that AI labs were not racing as fast as they can to train models that are much smarter than humans and unleash them on the world.
I wish that they hadn't already released models which generate images and audio that are hard to distinguish from real footage or the work of artists.
It is very clear that we could stop AI development right now and it would still transform a substantial portion of white collar work, often unrecognizably, over the next 5-10 years as people figure out how to make the technology work in various industries, even given current models' limitations.
I welcome Iceland to the EU if only so we might see Hlöllabátar in Europe. The continent needs Hlöllabátar.
Proud that after five years of careful analysis and discussion our paper "Lasting Lower Rhine-Meuse forager ancestry shaped Bell Beaker expansion" has now been published.
This paper in Management Science has been cited more than 6,000 times. Wall Street execs, top govt officials, and even a former U.S. Vice President have all referenced it. It’s fatally flawed, and the scholarly community refuses to do anything about it.
statmodeling.stat.columbia.edu/2026/01/22/a...
At the same time, posts from those two are likely responsible for a non-trivial proportion of the mania around genetic enhancement in Silicon Valley.
Blogging has consequences...
Fun that this study is substantially motivated by the 2013 "Lizardman's constant" post on Slate Star Codex. Like seeing those polygenic score benchmarking studies citing Gwern (as they should).
I think Trump's contribution to this decision is fairly small. EU membership has long been debated in Iceland and the two coalition parties pushing for a vote have basically been pro-EU since before Trump's first term. The upcoming referendum was agreed upon before Trump's second term began.
With most psychedelic drugs, you never know what you're going to get. But this mysterious mushroom from China - without fail - causes users to hallucinate tiny people: crawling up walls, popping out from under furniture and marching under doors. www.bbc.com/future/artic...
Brenna Henn Wanted to Improve Genetic Medicine. Then Her N.I.H. Grant Was Cut.
Paleogenomic insight into the collapse, recovery, and management of American bison https://www.biorxiv.org/content/10.64898/2025.12.24.696034v1
This smells distinctly like collider bias and/or selection bias and/or regression to the mean... You simply can't select teen prodigies, and world class athletes rom databases, and go run regressions without serious consideration of the selection process!
The U.S. Department of Justice has reached a $15 million NIH grant fraud settlement with Dana-Farber Cancer Center. This resolves a False Claims Act qui tam complaint that I filed on behalf of @sholtodavid.bsky.social in April 2024, with Gregg Shapiro as my co-counsel.
Sperm banks should be much more strictly regulated. (particularly, only accepting donations from registered donors). The banks don't seem to make the slightest effort to avoid these cases.
www.bbc.com/news/article...
Our new ancient DNA paper has just been published!
We present 28 new genomes from southern Africa - several of them high-coverage whole genomes.
Exciting to be moving towards population-level representation of ancient southern African genetic diversity!
www.nature.com/articles/s41...
Bayesian inference of population structure using identity-by-descent-based stochastic block models https://www.biorxiv.org/content/10.1101/2025.11.26.690810v1
This Kallmann diagnosis seems deeply implausible. For one thing he (famously!) was able to grow facial hair. And there's no indication that this intensely documented person had severely impaired smell, a distinctive feature of Kallmann. In fact, a little Googling specifically indicates the opposite:
I learn from my students of the Icelandic Saga map: hours of fun! sagamap.hi.is/is/
Revisiting the Evolution of Lactase Persistence: Insights from South Asian Genomes https://www.biorxiv.org/content/10.1101/2025.11.05.686799v1
Apropos of nothing in particular, I think it's useful to signal-boost this old paper on "how not to over-interpret ADMIXTURE bar plots". www.nature.com/articles/s41...
Can't believe I'd never heard of Euplotes before today: single-celled organisms who scurry around on little "legs"! Around 50 μm long, so ~1000x smaller than a tardigrade. This one has a kind of horseshoe crab habitus, flattened and with a "carapace". Incredible.
youtu.be/uJF30emkWt8?...
The big tech companies are reversing the intellectual and political gains of the Enlightenment and returning us to a New Dark Ages of ignorance, rage and superstition - it's the most important story of our time
jmarriott.substack.com/p/the-dawn-o...
Supervised mode after all just fixes some rows of the Q matrix to 1s and 0s and doesn't actually fix allele frequencies - P still gets updated given the whole data. Projection mode does literally fix (rather, overfit) the P matrix to the ref data, generally with poor results in my experience.
Indeed. But the inferred allele frequencies for a component (as given in the .P file) can get pulled away from the frequencies observed in its training pop and towards the freqs in the test samples if this allows better likelihood over the whole data, which can happen if test n >> training n.
Agree that different reference data is clearly warranted here, but if test n >> training n you would still likely see this kind of behaviour.
What I mean is that of the 5 superpopulations in 1000G, AMR has the lowest sample count. I have seen myself when test n >> training n that ADMIXTURE can basically give up on describing the ancestry of smaller training pops and instead use their maximised components to explain test sample variation.
Seems they used all 3,502 1000G individuals split by superpopulation as training samples. AMR is the smallest superpop. If test samples >> training samples (perhaps the case here?), it's easiest for ADMIXTURE to sacrifice the smallest training pop's ancestry component to describe test set ancestry
Something deeply strange happened to the ADMIXTURE run here! The American component, and to some extent the European one, seem to have been co-opted to describe Arab variation. Very little real indigenous American ancestry here, as indicated by e.g. mitochondrial haplogroup counts.
