iiinova — where your digital self keeps evolving

The preprint is available now, and we will also be presenting at AACL-IJCNLP in November! arxiv.org/abs/2603.21404

020

We show how multi-perspective LLM annotations can better support analyses when labels are subjective and vary across people.

131

These methods share a core assumption: that some notion of "ground truth" exists. For canonical tasks like digit recognition or cat-vs-dog classification, this is a reasonable assumption. But what about for more subjective tasks where human annotators may reasonably disagree?

There are a number of debiasing methods from the inference with predicted data literature (e.g. postpi, PP++, DSL, CDI) which show how you can still produce valid inference even when relying upon such flawed data.

120

LLMs are promising tools for reducing annotation burden in computational social science, but like outputs from any prediction model, they are imperfect proxies for human judgment.

Adam Visokay @avisokay.bsky.social · 30/09/2026

Looking forward to the Text as Data conference on Monday at UC Berkeley! I will be presenting our new paper "Multi-Perspective LLM Annotations for Valid Analyses in Subjective Tasks" which is a joint effort with Navya Mehrotra and @gligoric.bsky.social. 🧵

1224

Review so good I had to update the bio. (Rejected)

080

Reposted by Adam Visokay

naitian @naitian.org · 24/09/2026

If you haven't yet registered for TADA 2026 on October 5th (day before COLM), registration is open to the public! Detailed schedule forthcoming, but lots of fantastic posters and talks. We even have a second half-day of lightning talks scheduled! tada2026.eventbrite.com

tada2026.eventbrite.com

Text as Data 2026

A leading forum for interdisciplinary research on the study of politics, society, and culture through computational analysis of documents.

0177

Adam Visokay @avisokay.bsky.social · 24/09/2026

Ah okay cool! Well I might have some results to share from an ongoing project in the future!

111

Adam Visokay @avisokay.bsky.social · 23/09/2026

Great example. Are you aware of any literature evaluating this? RCTs or otherwise would be super helpful. Thanks!

111

Reposted by Adam Visokay

Hakeem Jefferson @hakeemjefferson.bsky.social · 15/08/2026

When I first went on the job market, I got offers from the best places & chose Stanford. The joy of that moment was tempered by the vitriol of those convinced I was unworthy of such opportunities. I learned long ago I’ll never convince them I am worthy. What great freedom came from that awareness.

438744

Adam Visokay @avisokay.bsky.social · 14/08/2026

epistemic risk machine ™️

000

Adam Visokay @avisokay.bsky.social · 12/08/2026

$\eta\epsilon\alpha\tau$ feature

020

Adam Visokay @avisokay.bsky.social · 12/08/2026

All of these projects look so great! Excited to read more closely

Adam Visokay @avisokay.bsky.social · 07/08/2026

How to Hide an Empire by Daniel Immerwahr is fantastic!

011

Adam Visokay @avisokay.bsky.social · 05/08/2026

But I think this is underutilized in sociology at the moment, and requires high quality validated data for calibration. Very sympathetic to having higher standards for reporting models/settings and also prompt engineering traces.

000

Adam Visokay @avisokay.bsky.social · 05/08/2026

I think you raise a really important point and I don’t know how we can meaningfully address the reproducibility concern. re LLM classification more broadly, there is great work which models prediction error and propagates it downstream (postpi, PPI, DSL). Lowers the stakes wrt exact LLM/settings.

100

Thank you!! Very excited for this chapter!

000

If you are interested in networks, multiverse methods, social construction, or economic sociology I would love to connect!

000

Next week in NYC I will be attending the @asanews.bsky.social annual meeting, presenting in the Market Codes: Bureaucrats, Brokers, and Algorithms (Mon 4PM) and New Advances in Social Network Analysis (Tue 10AM) sections. arxiv.org/abs/2603.00365 papers.ssrn.com/sol3/papers....

At Berkeley, I will also be working with Professor Joshua Blumenstock and @cega-uc.bsky.social on work at the intersection of development economics and computational social science.

100

This work has been shaped by my tremendous co-chairs Tyler McCormick and Sasha Johfre at the University of Washington!

In particular, I focus a lot on when we rely upon inexpensive proxies to make inferences about the more expensive to measure construct. This is increasingly common in this era of algorithms, predictions, and machine classification.

100

I'm excited to continue working on my dissertation project, Bodies of Knowledge, where I dig into the tensions that arise when the constructs we are interested in (e.g. body size) are operationalized through different measurement practices.

100