their terrible fake news lies, our righteous agitprop etc
11.03.2026 16:37
👍 5
🔁 0
💬 0
📌 0
their terrible fake news lies, our righteous agitprop etc
Concerning reports that Iran is attempting to mine the Strait of Hormuz. 1/
www.cbsnews.com/news/iran-mi...
i used this prompt exactly, you'll never guess what happened next
to think of LLMs creating, you must imagine a sculptor who can only cannibalize her own body to create.
model "prediction" is actually interpolation -- fitting a plausible next point on a learned curve. prediction carries intention. and it's interpolation within a fixed manifold, finding curves in the substrate no human could but unable to expand the manifold itself.
a mold doesn't "learn" the shape of what it produces; the resulting object can be structurally intricate & functional, but was not acquired through a learning process that reconfigured an existing entity. model architecture is inert; the learner doesn't exist prior to the learning.
imo importing terms like "training" & "prediction" from other parts of applied stats to describe attention-based transformers was a mistake. model params aren't records of training data the way memory is a record of experience. optimization is more like a mold imposing shapes on a material substrate
channeling my inner Mary Caprioli and coining the "gender shock" explanation for international conflict when a state experiences a sudden and dramatic change in Gender
it's very important that you keep resisting it
woke up this morning and decided it was my turn to build a Rust-based cli universal agent harness
The sky is not falling; high-quality platforms (Prolific, Verasight, CR Connect) have low rates of apparent bots. osf.io/preprints/ps... But also not zero; vigilance is very much needed!
i would love to know what their internal diagnostics and benchmarks are for emotional states like anxiety
the optimistic read on the old wars is that disciplines absorb disruption, fight it out, find some pragmatic middle ground. the less optimistic read is that the "middle ground" was cosmetic and the same people kept winning. which pattern do you think is going to repeat here
but more importantly old wars looked like they were about how to do research. LLMs sharpen the contrast of whether whole categories of intellectual labor just get automated. neither were truly methods debates, they were/are both labor market problems wearing a methods costume
also speed. the paradigm wars took three decades. ChatGPT hit a million users in a week and capabilities change every few months. the discipline's machinery for sorting out methods disputes runs on tenure-cycle time, it physically cannot keep up with the Anthropic release schedule
of course, the paradigm wars were a family fight. nobody outside the discipline gave a shit. the LLM thing involves a trillion dollar tech sector that needs you to adopt their product. that is a fundamentally different dynamic than arguing about KKV at APSA
the empirical picture is more mixed than we'd all care to admit. GPT-4 beats crowdworkers at annotation tasks by 25%, but switching the model you use can wildly swing treatment effects. we own none of the tools. it's an enormous replication crisis happening in plain view.
and now LLMs land right in the middle of this mess. the questions look the same, though. deskilling fears? we had those. black box anxiety? we had that. gatekeeping through journals/hiring? had that too. most grad students worry about skill atrophy *while still training*
and the resolution of all this was mostly aesthetic. "nominal pluralization," new journals, new APSA sections, rhetoric about methodological diversity -- all fairly surface-level accommodation, same power structure underneath
how many people remember the Perestroika email? Skocpol, Tilly, 200+ tenured faculty signing a petition against APSR becoming a quant monoculture. area studies people watching formal modelers fail to predict the Soviet collapse and get rewarded anyway. all things that really happened
besides some very recent work (eg Josh & Sol's linked Brookings piece) the debate over LLMs in political science has been bereft of recent history. we're working in the aftermath of a 30-yr war over methods that nearly crashed the discipline, after all :thread:
maybe so! i've been meaning to see it since so much was filmed in my home base of Albuquerque.
I'll be damned if that isn't a harbinger of things to come
Definitely that and a disinterest in/fear of the pedagogy of teaching research methods, which an outsider with a bit of pathos might salve