Lightnews — Scholar-powered news

Reposted by Michelle L. Ding

Valentina Pyatkin @valentinapy.bsky.social · 17h

💡We kicked off the SoLaR workshop at #COLM2025 with a great opinion talk by @michelleding.bsky.social & Jo Gasior Kavishe (joint work with @victorojewale.bsky.social and
@geomblog.bsky.social
) on "Testing LLMs in a sandbox isn't responsible. Focusing on community use and needs is."

1 4 14

Reposted by Michelle L. Ding

Vagrant Gautam @dippedrusk.com · Jun 11

Have you or a loved one been misgendered by an LLM? How can we evaluate LLMs for misgendering? Do different evaluation methods give consistent results?
Check out our preprint led by the newly minted Dr. @arjunsubgraph.bsky.social, and with Preethi Seshadri, Dietrich Klakow, Kai-Wei Chang, Yizhou Sun

Agree to Disagree? A Meta-Evaluation of LLM Misgendering

Numerous methods have been proposed to measure LLM misgendering, including probability-based evaluations (e.g., automatically with templatic sentences) and generation-based evaluations (e.g., with aut...

arxiv.org

1 4 15

Reposted by Michelle L. Ding

Elinor🎗️ @ COLM 🍁 @elinorpd.bsky.social · 23d

🚨 New preprint! 🚨
Excited to share my work: An AI-Powered Framework for Analyzing Collective Idea Evolution in Deliberative Assemblies 🤖🗳️

I’ll be presenting this at @colmweb.org in the NLP4Democracy workshop!

🔗 arxiv.org/abs/2509.12577

An AI-Powered Framework for Analyzing Collective Idea Evolution in Deliberative Assemblies

In an era of increasing societal fragmentation, political polarization, and erosion of public trust in institutions, representative deliberative assemblies are emerging as a promising democratic forum...

arxiv.org

1 2 7

Michelle L. Ding @michelleding.bsky.social · 1d

Read our full opinion abstract here: cntr.brown.edu/news/2025-09...

Testing LLMs in a sandbox isn’t responsible. Focusing on community use and needs is.

cntr.brown.edu

1 2

Michelle L. Ding @michelleding.bsky.social · 1d

Hi #COLM2025! 🇨🇦 I will be presenting a talk on the importance of community-driven LLM evaluations based on an opinion abstract I wrote with Jo Kavishe, @victorojewale.bsky.social and @geomblog.bsky.social tomorrow at 9:30am in 524b for solar-colm.github.io

Hope to see you there!

Third Workshop on Socially Responsible Language Modelling Research (SoLaR) 2025

COLM 2025 in-person Workshop, October 10th at the Palais des Congrès in Montreal, Canada

solar-colm.github.io

1 6 9

Reposted by Michelle L. Ding

Suresh Venkatasubramanian @geomblog.bsky.social · Jul 29

Very excited to be part of this new AI Institute that is being led by Ellie Pavlick @brown.edu and to be able to work with so many experts, including @datasociety.bsky.social

www.brown.edu/news/2025-07...

Brown University to lead national institute focused on intuitive, trustworthy AI assistants

A new institute, based at Brown and supported by a $20 million National Science Foundation grant, will convene researchers to guide development of a new generation of AI assistants for use in mental a...

www.brown.edu

2 4 23

Reposted by Michelle L. Ding

Serena Booth @reniebird.bsky.social · Jul 14

I'll be presenting a position paper about consumer protection and AI in the US at ICML. I have a surprisingly optimistic take: our legal structures are stronger than I anticipated when I went to work on this issue in Congress.

Is everything broken rn? Yes. Will it stay broken? That's on us.

A poster for the paper "Position: Strong Consumer Protection is an Inalienable Defense for AI Safety in the United States"

1 5 19

Reposted by Michelle L. Ding

Tech Policy Press @techpolicypress.bsky.social · Jul 7

With their 'Sovereignty as a Service' offerings, tech companies are encouraging the illusion of a race for sovereign control of AI while being the true powers behind the scenes, write Rui-Jie Yew, Kate Elizabeth Creasey, Suresh Venkatasubramanian.

'Sovereignty' Myth-Making in the AI Race | TechPolicy.Press

Tech companies stand to gain by encouraging the illusion of a race for 'sovereign' AI, write Rui-Jie Yew, Kate Elizabeth Creasey, Suresh Venkatasubramanian.

www.techpolicy.press

10 14

Reposted by Michelle L. Ding

Suresh Venkatasubramanian @geomblog.bsky.social · Jul 7

Very excited to see this piece out in @techpolicypress.bsky.social today. This was written together with @r-jy.bsky.social and Kate Elizabeth Creasey (a historian here at Brown), and calls out what we think is a scary and interesting rhetorical shift.

www.techpolicy.press/sovereignty-...

'Sovereignty' Myth-Making in the AI Race | TechPolicy.Press

Tech companies stand to gain by encouraging the illusion of a race for 'sovereign' AI, write Rui-Jie Yew, Kate Elizabeth Creasey, Suresh Venkatasubramanian.

www.techpolicy.press

9 21

Reposted by Michelle L. Ding

Suresh Venkatasubramanian @geomblog.bsky.social · Jun 12

So the EU AI Act passed. Companies have to comply. AI regulation is here to stay. Right? Right?

FAccT 2025 paper with @r-jy.bsky.social and Bill Marino (not on bsky) 📜 incoming! 1/n

arxiv.org/abs/2506.01931

Red Teaming AI Policy: A Taxonomy of Avoision and the EU AI Act

The shape of AI regulation is beginning to emerge, most prominently through the EU AI Act (the "AIA"). By 2027, the AIA will be in full effect, and firms are starting to adjust their behavior in light...

arxiv.org

2 4 9

Reposted by Michelle L. Ding

Victor Ojewale @victorojewale.bsky.social · Apr 28

Excited to present "Towards AI Accountability Infrastructure: Gaps and Opportunities in AI Audit Tooling" at #CHI2025 tomorrow(today)!

🗓 Tue, 29 Apr | 9:48–10:00 AM JST (Mon, 28 Apr | 8:48–9:00 PM ET)
📍 G401 (Pacifico North 4F)

📄 dl.acm.org/doi/10.1145/...

2 8 22

Michelle L. Ding @michelleding.bsky.social · Apr 26

all welcomes are good welcomes even with a 4 year delay 👍

2

Michelle L. Ding @michelleding.bsky.social · Apr 26

lol

Michelle L. Ding @michelleding.bsky.social · Apr 25

Also...super grateful & happy to be continuing my research journey at Brown in a CS PhD under @geomblog.bsky.social and @harinisuresh.bsky.social 🌱Many papers to come (?)

2 3

Michelle L. Ding @michelleding.bsky.social · Apr 25

yay!!

Reposted by Michelle L. Ding

Harini Suresh @harinisuresh.bsky.social · Apr 25

@michelleding.bsky.social has been doing amazing work laying out the complex landscape of "deepfake porn" and distilling the unique challenges in governing it. We hope this work informs future AI governance efforts to address the severe harms of this content - reach out to us to chat more!

Michelle L. Ding @michelleding.bsky.social · Apr 25

Excited to be presenting a new paper with @harinisuresh.bsky.social on the extremely critical topic of technical prevention/governance of adult AI generated non-consensual intimate images aka "deepfake pornography" at #CHI2025 chi-staig.github.io on 4/27 10:15-11:15 JST arxiv.org/abs/2504.17663 🧵

The Malicious Technical Ecosystem: Exposing Limitations in Technical Governance of AI-Generated Non-Consensual Intimate Images of Adults

In this paper, we adopt a survivor-centered approach to locate and dissect the role of sociotechnical AI governance in preventing AI-Generated Non-Consensual Intimate Images (AIG-NCII) of adults, coll...

arxiv.org

1 3

Michelle L. Ding @michelleding.bsky.social · Apr 25

Any synthetic content governance framework or method that aims to target AIG-NCII of adults should also consider their applicability to the malicious technical ecosystem and the three limitations we point out above. More during our presentation at chi-staig.github.io!

STAIG @ CHI 2025

chi-staig.github.io

1 2

Michelle L. Ding @michelleding.bsky.social · Apr 25

3. Adult AIG-NCII specific methods often focus on red teaming general purpose image generation models eg Stable Diffusion. While this is also important, there is an erroneous assumption of a "trustworthy technology" and "malicious users" when in fact the tech itself here is maliciously designed

1 2

Michelle L. Ding @michelleding.bsky.social · Apr 25

2. AIG-NCII prevention methods often conflate child sexual abuse materials (CSAM) with NCII of adults. But methods for CSAM (that often work with law enforcement databases) won't be as effective for adults due to different legal protections

1 2

Michelle L. Ding @michelleding.bsky.social · Apr 25

1. Transparency methods are very common in synthetic content governance frameworks but they are not enough. DF porn is not the same as other eg. political deepfakes bc human detectable images are still harmful. In fact, many of the deepfake creators label images as "fake" to avoid accountability

1 2

Michelle L. Ding @michelleding.bsky.social · Apr 25

Then we review the current landscape of synthetic content governance methods as recorded by NIST AI 100-4 nvlpubs.nist.gov/nistpubs/ai/... and show 3 key limitations in governing AIG-NCII of adults

nvlpubs.nist.gov

1 2

Michelle L. Ding @michelleding.bsky.social · Apr 25

Technical prevention is then a challenge for sociotechnical AI governance. In our paper, we break down and map what we call the "malicious technical ecosystem" that is used to create AIG-NCII of adults, including open-source face swapping models and 200+ "nudifier" apps that are free & easy to use

1 2

Michelle L. Ding @michelleding.bsky.social · Apr 25

There is a lot of work on responding to AIG-NCII through improved take down mechanisms (eg Take It Down Act) and legal recourse for survivors (eg DEFIANCE Act), but response without prevention places the burden of removal and justice-seeking on survivors and does nothing to stop the creation of NCII

1 2

Michelle L. Ding @michelleding.bsky.social · Apr 25

AI generated NCII is a form of image based sexual abuse that results in severe mental, physical, financial, and reputational damage as well as a gendered chilling effect. myimagemychoice.org is an organization that documents this extensively through survivor testimonials

Home - #MyImageMyChoice

myimagemychoice.org

1 2

Michelle L. Ding @michelleding.bsky.social · Apr 25

Excited to be presenting a new paper with @harinisuresh.bsky.social on the extremely critical topic of technical prevention/governance of adult AI generated non-consensual intimate images aka "deepfake pornography" at #CHI2025 chi-staig.github.io on 4/27 10:15-11:15 JST arxiv.org/abs/2504.17663 🧵

The Malicious Technical Ecosystem: Exposing Limitations in Technical Governance of AI-Generated Non-Consensual Intimate Images of Adults

In this paper, we adopt a survivor-centered approach to locate and dissect the role of sociotechnical AI governance in preventing AI-Generated Non-Consensual Intimate Images (AIG-NCII) of adults, coll...

arxiv.org

1 2 7