Yes yes
05.03.2026 22:03
👍 3
🔁 0
💬 0
📌 0
Yes yes
why can't we bash both
(FWIW this answer is also why I don't understand people dunking on sean westwood for his point on lit reviews...)
yes AI cannot make you an expert in the paper. if you have to write and do a full 40 minute job market style talk on it, that'll catch AI use most of the time IMO
then, i have another agent be the antagonist/reviewer 2 to all answers generated, and iterate over this. this would work remarkably well and probably better than most people who spent months studying coudl do themselves 2/2
before people say "well LLMs hallucinate and are bad writers":
say i have a reading list for comps. I have zotero export ALL papers from that list (plus some other core papers I think are necessary) into a folder. I have CC ingest those papers and tell it to ONLY use those papers for answers 1/2
absolutely not useful
an important and underdiscussed point on the rise of agentic AI
Lol sold
yeah this is exactly my take. in crim it's particularly pernicious because the results of these "quasi experimental" designs are aggregated up (sometimes explicitly) to say Policy X works, this is well known, you must use policy X in your police department.
in leading criminology journals (such as the Journal of Experimental Criminology) "quasi-experimental" is frequently used in explicitly non-causal designs, such as purely a selection on observables conditional independence case (we controlled for all the differences). this seems bad to me.
it actually might be pretty good at that lol
or it's best used when vibe coding some bullshit app lol (truly incredible on this front)
when you read the phrase "quasi-experimental" what does it mean to you?
I would change this slightly to say I think they are best used in settings where you yourself have expertise
interesting! i'm not aware of this literature, but surely the aggregated effect entirely depends on what the constellation of heterogenous effects are?
if, like me, you were not aware of this: Ghost Criminology is a thing (but it's also not real science):
sean is right about this, but it's not clear what things will look like which is why it's hard to talk about concretely
no i agree with that, but it's been clear for over a year now online assignments are dead. more generally regarding online courses, i'm having a hard time envisioning a future where employers see those as valuable at all
i mean, if you're using canvas for assignments these days and this is a shock to you i don't know whta to tell you
i mean very possible i'm over fitting based on my personal experience & workflow. I have tons of projects with vastly different datasets and this makes me able to actually pursue a lot more of them in a serious way
yes i agree with this, though i could pull this off -- it'd take me longer -- and i have nowhere near the experience he does.
what i'm saying is for researchers it magnifies experience by 5-10x or whatever arbitrary number. thus if your'e not using it, you're not benefiting from that magnification
i'm nto sure why you are so antagonistic towards me on this.
autonomy is not the point, is all i'm saying
i don't think any of this has to do with autonomous
it requires that amount of background *right now*
Yes sorry not the reproducibility thing per se, but this as an example of how powerful these tools are for our work
This also demonstrates the great strides political science has made towards transparency and reproducibility. Such a task would be literally impossible in the criminology discipline, for instance.
An astounding piece of work, full thread here: x.com/xuyiqing/sta...
I have not seen this mentioned here, but this is a great example of how the research world has dramatically changed underneath of us and if you're not keeping up you're going to be left behind quickly:
yiqingxu.org/papers/2026_...
@mattgrossmann.bsky.social, @joshmccrain.bsky.social, and I are putting together a panel at PolMeth in July on using CongressData, CSPP, and SenateData. We are aiming to have up to 5 people (preference to junior faculty using the data or interested in doing so). Please reach out to me if interested!