iiinova — where your digital self keeps evolving

impactful-pluralistic-alignment.github.io

Huge thanks to my amazing coauthors 🙏 Jillian Fisher @atoosakz.bsky.social Jacob Andreas @mitchellgordon.bsky.social @mbakker.bsky.social , and shoutout to @taylor-sorensen.bsky.social for all the helpful feedback! Paper: impactful-pluralistic-alignment.github.io 10/10

A Roadmap to Impactful Pluralistic Alignment Research

030

Adoption will ultimately need frontier labs' cooperation. But much of the gap is ours (researchers) to close: making the empirical case, settling what ideal pluralistic behavior looks like, and building evals + methods feasible for deployment. The roadmap is in the paper 🗺️ 9/

We consider the strongest objections: Shouldn't regulation handle this? What about open models🐳? Might pluralism already emerge on its own? Is this the wrong goal entirely🎯? We respond to 8 alternative views. Most, taken seriously, support our same call to action. 8/

3: evals + methods aren't adoptable today. Existing metrics aren't hill-climbable, eg coverage scores reward adding perspectives even when it hurts accuracy or usefulness, and tradeoffs go unmeasured We need to build towards release-ready evals + methods fit for deployment. 7/

2: we haven’t settled when pluralism is warranted or what it looks like ideally. "Subjective" queries are too vague. Research can inform how user+societal context *should* influence operationalization. We need a concrete goal that a model spec could state. 6/

So why no adoption? Reason 1: the case for pluralism is mostly normative. No study has tested whether pluralistic responses actually benefit users: depolarization, epistemic autonomy, representation, etc. We call for human studies measuring benefits and costs together. 5

Evaluations are where priorities show. No frontier lab publicly focuses on pluralism model card evaluations The one dedicated pluralism eval ever published (Muse Spark) measured individual preference prediction, and was dropped again for the 1.1 release. 4/

We audit the public docs of the most widely used AI models: Anthropic, OpenAI, Google, xAI, Meta (+ top open-weight families) In behavior docs: 4 of 5 prescribe some multi-perspective behavior. None name pluralism as a goal or indicate training for it explicitly 3/

To me, working on this shouldn't be just an academic pursuit. Pluralistic alignment aims to solve important problems in the real world, affecting billions of humans worldwide🌍 So our ask is collective: let's prioritize research that builds towards deployed impact 💪 2/

David Jurgens @davidjurgens.bsky.social · 13/07/2026

Pluralistic alignment is thriving as a research agenda yet failing at its goal: making the AI systems people actually use more pluralistic🌈 🚨New position paper: we argue adoption in deployed models should be the fields main goal & we provide a roadmap of how to get there 🧵1/

3212

Reposted by Elinor

With my ACL PC abilities, I wanted to characterize who these folks are, whether they're actually "outsiders", whether their papers are much less likely to be accepted, is this growth just LLM slop? The whole analsis is here medium.com/@jurgens_245...

medium.com

Is the ACL Rolling Review actually broken?

ARR is broken! ARR is being overwhelmed with papers from outsiders! LLM-generated slop is ruining our peer review! There are not enough…

12811

Reposted by Elinor

Jonathan Stray @jonathanstray.com · 23/10/2025

GreenEarth is creating open source AI-driven recommender infrastructure for BlueSky. Type a prompt, see your feed change. We are here for the users, the builders, the dreamers. Join us. greenearthsocial.substack.com/p/introducin...

greenearthsocial.substack.com

Introducing GreenEarth

We're building advanced open source algorithms for social media

1010427

Reposted by Elinor

iNaturalist @inaturalist.bsky.social · 06/07/2026

We're looking for a CV/ML Engineer to help us improve the machine learning systems that power iNaturalist's species identification and geographic range modeling. If you're excited to help build tools that help millions of people engage with nature, we'd love to hear from you! Apply: buff.ly/YZqaW6c

Image of flowers with text overlaid saying: "iNaturalist: We're hiring! Computer vision/machine learning engineer. Full time, remote in the United States."

03921

Elinor @elinorpd.bsky.social · 14/06/2026

Just uploaded a new plant blog post on my website 🥰🪴🌸

270

Reposted by Elinor

Ehud Reiter @ehudreiter.bsky.social · 08/06/2026

New blog: I am worried by NLP research culture NLG and NLP are mostly much better in 2026 than when I got my PhD in 1990. Unfortunately research culture has gotten *worse” in this period, which really worries me as I retire. ehudreiter.com/2026/06/08/n...

ehudreiter.com

I am worried by NLP research culture

In most ways NLG and NLP are much better in 2026 than when I got my PhD in 1990. Unfortunately research culture has gotten *worse” in this period, which really worries me as I retire. We have…

1144

Elinor @elinorpd.bsky.social · 06/06/2026

thank you! (and thanks to them!)

000

Elinor @elinorpd.bsky.social · 06/06/2026

excited to share that i'll be pursuing my phd in computer science at @csail.mit.edu starting this fall 🥳 🎓 i'm so grateful to be coadvised by the literal dream team: Jacob Andreas, @mbakker.bsky.social, and @mitchellgordon.bsky.social 🙌

4240

Reposted by Elinor

Benno Krojer @bennokrojer.bsky.social · 05/06/2026

My first last-author paper is out! If you saw this dog below and someone showed you the second image, would you consider them the same word/concept? (more examples in Ada's thread) We study if VLMs agree with humans on this and revisit old questions around shape vs. texture bias in vision

1124

Reposted by Elinor

Jessica Hullman @jessicahullman.bsky.social · 28/05/2026

Thoughts on metascientific consequences of AI-generated slides & ideas diluting the impression that speakers are commited to what they present. Science runs on personal attachment more than we admit. If it were a cake mix, how wouldn we add back an egg? statmodeling.stat.columbia.edu/2026/05/28/w...

statmodeling.stat.columbia.edu

What if scientists really were dispassionate observers, communicating ideas without irrational commitment? Look here, says AI. | Statistical Modeling, Causal Inference, and Social Science

1296

Reposted by Elinor

Ilias Chalkidis @kiddothe2b.bsky.social · 06/05/2026

📢 Our paper 🤯🧠 "Brainrot: Deskilling and Addiction are Overlooked AI Risks" has been accepted at the ACM Fairness, Accountability & Transparency (FAccT) conference 2026. The preprint is available: 👉 arxiv.org/abs/2605.03512 TL;DR 🧵 follows 👇 1/5

arxiv.org

Brainrot: Deskilling and Addiction are Overlooked AI Risks

The scope of AI safety and alignment work in generative artificial intelligence (GenAI) has so far mostly been limited to harms related to: (a) discrimination and hate speech, (b) harmful/inappropriat...

1215

Reposted by Elinor

Joachim Baumann @joachimbaumann.bsky.social · 01/05/2026

Can you boost your AI review scores by asking an LLM to rewrite your paper? Yes! We call it paper laundering Our @icmlconf.bsky.social spotlight paper argues current AI reviewers aren't ready to automate peer review, and outlines what a science of peer review automation should look like 🧵👇 #ICML2026

First page of the ICML 2026 spotlight paper "Stop Automating Peer Review Without Rigorous Evaluation" by Joachim Baumann, Jiaxin Pei, Sanmi Koyejo, and Dirk Hovy (Stanford University and Bocconi University). The abstract argues that today's AI systems should not be used to produce paper reviews, grounded in two empirical findings: a "hivemind effect" where AI reviewers show excessive agreement and reduce perspective diversity, and "paper laundering," where prompting an LLM to rewrite a paper trivially increases AI reviewer scores through stylistic changes rather than scientific improvements. The paper calls for a science of peer review automation rather than wholesale deployment of general-purpose LLMs.

44314

Reposted by Elinor

Petter Törnberg @pettertornberg.com · 01/05/2026

LLMs have been widely reported as left-wing biased. The finding has shaped policy and debate — with Trump banning "Woke AI". Our new paper challenges this story. It's not that the models are biased. It's that they think the auditor is. 🧵 w/ michelleschimmel.bsky.social‬arxiv.org/pdf/2604.27633

415750

Reposted by Elinor

Shahan Ali Memon @shahanmemon.bsky.social · 30/04/2026

Our position paper, "AI Welfare Is Bullshit" just got accepted to ICML 2026 @icmlconf.bsky.social! The AI welfare agenda has already begun to attract institutional investment at organizations such an @anthropic.com. We argue that this idea is essentially Frankfurtian Bullshit.

56219

Reposted by Elinor

Mark Riedl @markriedl.bsky.social · 30/04/2026

Designating that most uses of the term "goblin" and "gremlin" are not legitimate while designating most uses of the word "frog" as legitimate is the imposition of a set of values on a technology. Why do a small number of people in Silicon Valley get to decide whether goblins are inappropriate?

5736

Reposted by Elinor

Myra Cheng @myra.bsky.social · 23/04/2026

Patrick Sui and I are hosting an #ICLR2026 social for anyone with background/interest in the humanities! Room 210, 12-1:30pm on Friday 24 April!! Humanities-adjacent, humanities-curious, everyone is welcome! Should be a fun group! :)

1217

Elinor @elinorpd.bsky.social · 22/04/2026

Flying out of Boston to Brazil rn means I’m surrounded by poster tubes (ICLR attendees) and blue athletic gear w medals (Boston marathoners). It’s a cool crowd

050

Elinor @elinorpd.bsky.social · 21/04/2026

I'll be presenting OvertonBench at #ICLR2026 in Rio later this week! 📍Sat, Apr 25, 10:30am in Pavilion 4 (#4109) Please DM me if you'd like to chat about pluralistic / value alignment, societal impacts, epistemology, fairness, evals, etc

081

Reposted by Elinor

MIT Siegel Family Quest for Intelligence @mit-sqi.bsky.social · 17/04/2026

Congratulations to Jacob Andreas, was named a 2026 Edgerton Award recipient! The award recognizes exceptional teaching, research, and service at MIT! Prof. Andreas co-leads our Language and Thought Mission, and he is a dedicated and creative researcher and educator. news.mit.edu/2026/jacob-a...

news.mit.edu

Jacob Andreas and Brett McGuire named Edgerton Award winners

MIT associate professors Jacob Andreas and Brett McGuire have been selected as the winners of the 2026 Harold E. Edgerton Faculty Achievement Award for exceptional contributions to teaching, research,...

171

Elinor @elinorpd.bsky.social · 18/04/2026

My Hoya Rebecca put out the cutest teeny lil flowers today and I’m obsessed

030

Reposted by Elinor

David Bau @davidbau.bsky.social · 08/04/2026

Tech industry mottos have a mixed track record. But we should hold idealists to their ideals. And we should celebrate when they come through. The Mythos non-release is a remarkable moment of conviction. Thoughts: davidbau.com/archives/20... Bravo to Anthropic's "race the top".

1133

Reposted by Elinor

Gillian Hadfield @ghadfield.bsky.social · 03/04/2026

Democracy isn't a rulebook. It runs on daily interactions where people comply with norms and hold each other accountable. AI agents are about to join that system. We need to build them to read it. New paper with Rakshit Trivedi and Dylan Hadfield-Menell.

knightcolumbia.org

Building AI for the Democratic Matrix: A Technical Research Agenda for Normative Competence and Normative Institutions

To maintain democratic resilience, it is essential to build AI agents capable of choosing behaviors that mirror those of the human agents that constitute human democracies.

4123

Reposted by Elinor

Leshem (Legend) Choshen @EMNLP @lchoshen.bsky.social · 30/03/2026

Rebuttal season is here, yey🤞 With many asking me, I compiled the most common misconceptions Hope the tips help 🧵 All tips: docs.google.com/document/d/14Wax8M5w8F_8miDlYJ9-I6wqpelxlXjCEUbkNzNMqqE/edit?tab=t.0#heading=h.rfq27f356vmm #AI 🤖📈🧠

152

Reposted by Elinor

Mor Naaman @informor.bsky.social · 23/03/2026

New important (I hope) resource for academics working in this area.

191

Reposted by Elinor

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 15/03/2026

For a recent lab meeting, I wrote up a grab bag of ways to think about your development as a researcher during a PhD: emerge-lab.github.io/papers/an-un... Sharing in case folks find it useful or have feedback!

emerge-lab.github.io

610112

Elinor @elinorpd.bsky.social · 12/03/2026

its worthwhile for future work to actually understand which types of queries we want Overton pluralistic responses (& how it relates to human prefs) Using the metric to understand models pluralistic capabilities vs directly as a reward/optimization are different, its not designed for the latter rn

020

Elinor @elinorpd.bsky.social · 12/03/2026

The paper frames higher as better but doesn’t assert that all models should strive for perfect scores all the time. And the *subjective* queries on which we want any Overton pluralism is pretty narrow. The past example is rlly interesting though! Hints at a general problem of Overton Window shifts

Mor Naaman @informor.bsky.social · 11/03/2026

Reposted by Elinor

🥁🥁🥁 Newly out from us today in Science Advances: “Biased AI Writing Assistants Shift Users’ Attitudes on Societal Issues”. Large Language Models are providing users with autocomplete writing suggestions on many platforms. Could these suggestions shift users’ own attitudes? (spoiler: YES) (1/7)

5195105

Elinor @elinorpd.bsky.social · 11/03/2026

the key points are 1) its important to measure pluralistic capability bc we think its necessary for better overall value alignment 2) its especially important to understand its impacts on users (future work) + tradeoffs wrt political neutrality, etc