Lightnews — Scholar-powered news

Alexander Hoyle @alexanderhoyle.bsky.social · 21h

I would like the full slide deck!

Reposted by Alexander Hoyle

Manoel Horta Ribeiro @manoelhortaribeiro.bsky.social · 3d

Computer Science is no longer just about building systems or proving theorems--it's about observation and experiments.

In my latest blog post, I argue it’s time we had our own "Econometrics," a discipline devoted to empirical rigor.

doomscrollingbabel.manoel.xyz/p/the-missin...

2 9 29

Alexander Hoyle @alexanderhoyle.bsky.social · 2d

really like this post! I feel that ML/NLP is an often empirical field that fails to adopt the practices of one—and being in an econ group for my postdoc, I’ve also noticed the big gulf in rigor. curious to see how your course shapes up (you might be interested in this paper arxiv.org/abs/2411.10939)

2

Alexander Hoyle @alexanderhoyle.bsky.social · 9d

Circular meaning: if you model bargaining behavior using personas with different risk profiles and recover behavior that mirrors those risk profiles, what have you learned?

1

Alexander Hoyle @alexanderhoyle.bsky.social · 9d

Nice share I missed before! I've been thinking of an similar argument along these lines. We lack a good mathematical model of psychology undergraduates in a lab, so to imagine we can simulate whole populations via prompting seems like wishful thinking (not to mention potentially circular)

2 1

Alexander Hoyle @alexanderhoyle.bsky.social · 9d

You're doing terrific work. In fact, I'll be referencing your "Using ChatGPT is not bad for the environment" research in a talk tomorrow

1 4

Alexander Hoyle @alexanderhoyle.bsky.social · 13d

Was about to share this with you @nbalepur.bsky.social until I realized it was your paper :P

1 2

Alexander Hoyle @alexanderhoyle.bsky.social · 14d

Thanks ! also, we should have a zoom catchup soon !

1

Alexander Hoyle @alexanderhoyle.bsky.social · 14d

Accepted to EMNLP (and more to come 👀)! The camera ready version is now online---very happy with how this turned out

arxiv.org/abs/2507.01234

Alexander Hoyle @alexanderhoyle.bsky.social · Jul 17

New preprint! Have you ever tried to cluster text embeddings from different sources, but the clusters just reproduce the sources? Or attempted to retrieve similar documents across multiple languages, and even multilingual embeddings return items in the same language?

Turns out there's an easy fix🧵

Barchart of number of items in four clusters of text embeddings, with colors showing the distribution of sources in each cluster.

Caption: Clustering text embeddings from disparate sources (here, U.S. congressional bill summaries and senators’ tweets) can produce clusters where one source dominates (Panel A). Using linear erasure to remove the source information produces more evenly balanced clusters that maintain semantic coherence (Panel B; sampled items relate to immigration). Four random clusters of k-means shown (k=25), trained on a combined 5,000 samples from each dataset

4 12

Alexander Hoyle @alexanderhoyle.bsky.social · 14d

? I didn't delete anything

1 1

Alexander Hoyle @alexanderhoyle.bsky.social · 14d

You might be interested in this blog post, which discusses these issues in depth: andymasley.substack.com/p/individual...

1 4

Alexander Hoyle @alexanderhoyle.bsky.social · 16d

This is terrific advice (that I'd do better to take into account myself, too). Whenever I've read more deeply into a subject/theory, it ends up paying dividends

Alexander Hoyle @alexanderhoyle.bsky.social · 20d

hot take: I think the one at Kebab aur Sharab on the UWS is better (although i didn’t go to the main dishoom location last time I was in London and I think their QC maybe was not up to par?). Also terrific butter chicken

1

Alexander Hoyle @alexanderhoyle.bsky.social · 21d

Yes—in fact a couple years ago I’d been chatting with your now-colleague Maria Pacheco about a paper with the premise “the methods are new but the the pitfalls remain” re: CSS/TADA research“

3

Alexander Hoyle @alexanderhoyle.bsky.social · 23d

@mariaa.bsky.social I think you'll appreciate it

1 3

Alexander Hoyle @alexanderhoyle.bsky.social · 23d

this looks terrific, very excited to read

Ethan Mollick @emollick.bsky.social · 23d

LLMs introduce a huge range of new capabilities for research, but also make it possible for researchers to "hack" their results in new ways by how they chose to use models for annotation

This is a useful pass at quantifying some of the risk, and some mitigation strategies arxiv.org/pdf/2509.08825

1 6

Alexander Hoyle @alexanderhoyle.bsky.social · Aug 27

Spoiler alert!! I’m currently reading A Place of Greater Safety

1 2

Alexander Hoyle @alexanderhoyle.bsky.social · Aug 24

You swap between these regularly? Or is there a way to consolidate ? Staying on top of my half dozen active slacks is already too much

2 2

Alexander Hoyle @alexanderhoyle.bsky.social · Aug 15

a friend bought these a size too large and I love them . Can wear to work or (as I just did) on a long haul flight

1

Alexander Hoyle @alexanderhoyle.bsky.social · Aug 13

While writing my dissertation, I came to realize we perhaps presented an overly neat view of QCA in the second paper (eg, that inter-annotator agreement is an unalloyed good) and failed to characterize the full extent of practices/debates. But I still think an unstable model is not desirable :)

1

Alexander Hoyle @alexanderhoyle.bsky.social · Aug 13

Of course! we also have two others showing LDA is better than some newer models, but not BERTopic specifically:

aclanthology.org/2022.finding...
dl.acm.org/doi/10.5555/...

Also: been watching your IC2S2 keynote and am learning a lot!

Are Neural Topic Models Broken?

Alexander Miserlis Hoyle, Rupak Sarkar, Pranav Goel, Philip Resnik. Findings of the Association for Computational Linguistics: EMNLP 2022. 2022.

aclanthology.org

1 3

Alexander Hoyle @alexanderhoyle.bsky.social · Aug 13

Yep, unfortunately that can be an issue---if you need support, these all show BERTopic underperforming (apologies for the blatant self-promotion):

aclanthology.org/2025.acl-lon...
aclanthology.org/2025.acl-lon...
aclanthology.org/2024.naacl-l...
aclanthology.org/2024.eacl-lo...

1 5

Alexander Hoyle @alexanderhoyle.bsky.social · Aug 13

One piece of advice (which applies to all topic models) is to vary hyperparameters (like the underlying sentence embedding model; min cluster size) and see if the outputs differ

2

Alexander Hoyle @alexanderhoyle.bsky.social · Aug 13

Yeah, we've found BERTopic is empirically worse than bog-standard LDA (Mallet/Tomotopy) across several different evaluation settings. It also is less stable

It can be useful if you have multilingual data, although other options like CTM may be preferable. Is there a reason you need to use BERTopic?

2 2

Reposted by Alexander Hoyle

Dallas Card @dallascard.bsky.social · Jul 29

I am delighted to share our new #PNAS paper, with @grvkamath.bsky.social @msonderegger.bsky.social and @sivareddyg.bsky.social, on whether age matters for the adoption of new meanings. That is, as words change meaning, does the rate of adoption vary across generations? www.pnas.org/doi/epdf/10....

3 13 48