PDF candidate @PSUSocCrim & @CSoDA_PSU | (Culture Politics Sci. of Sci.) x Computational Social Sci. | @sehiruniversite & @BingPoliSci Alum |
Joined February 2012
- Tweets 11,155
- Following 321
- Followers 1,221
- Likes 15,485
1,333 Photos and videos
Pinned Tweet
23 Oct 2024
Spite. Pure spite!
23 Oct 2024
How do you stay excited about doing research?
1
4
1,087
This is how good the best LLMs are at identifying conspicuous ideological bias (Allsides), let alone identifying bias in academic texts.
For reference, the author of that paper uses GPT-4, which is objectively worse than the least performing LLM in the list 🤗
Jun 15
"anglophone social science abstracts have been persistently classified to the left of center for decades, and this asymmetry has intensified over the past thirty years, most sharply on sociocultural dimensions." link.springer.com/article/10…
1
227
Wrote about the selection bias in the data itself. Without any true labels of human annotated data to judge and comparing it with performance of 10x better state of the art LLMs, the LLM used in that paper did as good as a coinflip.
x.com/i/status/2034431569901…
Let's understand how abysmal the data and methods are, and how peer review is being butchered at Theory and Society for ideological hit-pieces.
80% of OpenAlex pubs have no abstract (copyright). E.g., only 5.6% of ASR entries havd abstracts.
1
173
Going to replicate it in full whenever I get time and money to get people to stop pouring LLMs on sensitive tasks (like ideological classification) like ketchup 🤗
47
From the first decision, to reviews, to revise and resubmit, and publication, ~10 days 😊
Very disappointed to see that @AramHur's latest paper in @koreaobserver has six completely fabricated citations.
Academic integrity matters, people.
A short 🧵
362
OK, what does reform look like?
Forcing people to work on things they would otherwise not work on voluntarily?
DEI, but of the “right” type?
That said, every single sociologist is aware of the ideological skew. Graduate students are actually told to tone down their “Marxism”.
Jun 9
Sociologists have a choice.
1. Admit their field is devoted to left-wing activism. Be proud of it, openly defend it, and let the public decide if they want to support it. Don't throw a fit if they say no.
2. Admit it and demand reform.
3. Deny it and blame people for noticing.
2
540
Try submitting an activist paper to the mainstream journals i.e., AJS, ASR, SF, and see if it gets published. Chances are close to none.
The specter of the Left is actually American pragmatism and progressivism. Most academics are just publishing or perishing without believing.
146
OK, me and my friends think everybody is a Marxist, writing non-peer reviewed opinion pieces with no concrete evidence but only anecdotes.
So, we too wrote a non-peer reviewed opinion piece with no concrete evidence but only anecdotes. Fighting fire with fire 🤗
1
161
“Isn't it obvious” isn't strong evidence.
I have yet to encounter a hardcore leftie in Soc and Polisci deptts I've been too. Everyone is, at best, a normie, going with the flow.
I've been warned though not to cite Marx, Gramsci, Foucault etc., every single time I cited them!
1
136
Very few faculty in mainstream soc sci departments are lefties. They may vote for Democrats, but for anyone coming from outside the US, their politics is center-right at best.
The social justice warrior class is usually excluded from the mainstream deptts and journals.
89
Abdul Basit Adeel et al. retweeted
May 28
Thrilled to share our new paper, "What Would Jesus Fund? Christian Nationalism and Americans' Views on Government Spending," just out in Socius! 🧵
journals.sagepub.com/doi/10.…
1
1
8
368
Most qualified person Fareed found to discuss democratic backsliding in Turkey is neither Turkish nor an academic, but a travel writer, who in between the lines repeats the trope that Erdoğan took refugees in based on some sort of a plan (like they had somewhere else to go?).
May 24
Turkey has descended down the path of illiberal democracy over the past few decades. I talked to journalist @suzyhans about her new book that observes one neighborhood in Istanbul over a decade to try to understand the changes taking place under President Erdoğan:
1
243
The idea that Erdoğan “imported” refugees based on some clever plan to change Turkey's politics through demographic engineering is a dangerous trope. Turkey had the capacity to absorb them then (not now).
Plus, Lebanon and Jordan received more per capita refugees than Turkey.
1
60
Interesting study; massive under-counting.
- Misses 68% of the True Positives (recall).
- O3 is a good model not great; definitely missing even a lot more cases.
That's passage-level stats; whole sermon classification is definitely less precise. Needle in the haystack problem.
How common is political advocacy from the pulpit? We find it is both more common and more geographically dispersed than previously understood. The American pulpit now functions as a significant channel for partisan base mobilization.
pnas.org/doi/10.1073/pnas.26…
75
Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect. Selection effect.
"Relative to the general population, liberals are overrepresented among faculty by 186% - vastly more than men and considerably more than Asians. Meanwhile, conservatives are underrepresented by 70% - more than Hispanics, Black people, and Native Americans."
stevestewartwilliams.com/p/a…
101
Artists can tell whether a song is AI generated in seconds because it has that “feel”.
So do academics when the text is written by AI. There is no mathematical formula, but the text shouts it is AI written.
Don't do it. Don't let your reviewers suffer!
1
90
And definitely don't claim, “I “read” and “classified” ~10,000 texts into 26 distinct labels”. That takes at least a year with a team of RAs!
You would know if you actually did that 😘 Read a 1,000 and compare LLMs performance against them. More than enough 😘
37
Yeah, because everyone is using LLMs to write code and pass it on as a package they "created". Witnessed at least two presentations like that.
Basically AI slop coming to open-source stats packages = more replication crises 😘
Indeed. No evidence of precipitous declines in package and repo creation (relative to R, even).
Thread on this, using CC, here: x.com/akib_kn/status/2049138…
2
304
So, basically DEI for folks who (not all, but most) believe DEI has/d zero net benefits 😊
Note: not insinuating that my friend Jesse is one of those. Instead he is making a valid intellectual point, yet the irony of the situation needs to be pointed out.
Apr 22
Nice summary case for the *intellectual* (as opposed to political) benefits of getting more conservatives in higher ed. Similar to arguments I’ve been making but with that Harvard sheen! thecrimson.com/article/2026/…
1
136
DEI, with all its problems and politicization, was intended to bring un/under-represented voices and experiences to academia.
Conservatives are now excluded from academia, so such calls are basically DEI for conservatives. Different group, same logic, and good logic.
1
71
Dayının da dediği gibi “Ortadoğunun kralıyız ...” 😘
Apr 8
Turkey is not in the Middle East
196