Statistical geneticist | Associate Prof at @DanaFarber / @harvardmed / @DFCIPopSci | Blogging at theinfinitesimal.substack.co…
- Tweets 23,423
- Following 2,929
- Followers 25,402
- Likes 23,884
Pinned Tweet
I've written about race, genetic ancestry, analyses of large biobanks, and human history
gusevlab.org/projects/hsq/#h…
I'll summarize the key points here 🧵:
51
313
1,388
635,931
We have a new paper on topic models!
We introduce Mechanistic Topic Models (MTMs). MTMs model SAE feature counts rather than words or plain embeddings, combining the benefits of LLM-based and probabilistic topic models.
Paper (published in TACL) : arxiv.org/abs/2507.23220
6
14
124
6,859
Me, to AI: "Wow, we just incorporate a lot of new biological data and the performance decreased. That's very surprising! What do you recommend we do next?"
AI: "We should change our file format to include more metadata and a fingerprint of the run parameters"
6
1
68
7,924
Hard utilitarianism seems to be attempting to develop a kind of central planning for moral decisions (I realize this is not a new observation). And I suspect that's why super accurate AI superforecasters are such an appealing idea for some utilitarians.
5
1
47
3,219
Why would moral value be predictable to begin with and not be swamped by aleatoric uncertainty? Well, IQ is correlated between MZ twins at about the same level as it is within a retested individual (which is to say, highly). And IQ is a defining characteristic of moral worth.
1
10
1,034
This thread is an interesting discussion of data gatekeeping in behavior genetics. One thing that is noticeably absent is any consideration for the *participants* in the underlying studies. This issue is almost always presented as a Manichean battle between the do-gooder ...
Undoubtedly. Unfortunately the current gatekeepers and censors of behavioural genetics probably consider themselves "sophisticated" consequentialists because they are trying to prevent "indirect" social consequences. But they are failing to understand the adverse second-order effects of infringing on academic freedom and cannot conceive that it might be better if society knew the truth about controversial topics rather than remaining in a state of ignorance.
5
12
78
17,577
So that has left the field with about a decade of available data full of stratified garbage, often leveraged for pseudoscientific ends; the methods to clean out the garbage; and no interest from industry to actually enable the access needed to do so.
1
23
1,497