Joshua Ashkinaze

PhD candidate @UMSI. Interested in computational social science / HCI /multi-agent systems / alignment. Also into mathematical art (p5.js, processing)

[Joshua Ashkinaze @JoshAshkinaze]

3/ To test this, we use a confound-then-deconfound design. In "training," a user's past choices perfectly confound a deep value with a shallow preference. At test, we break that link and see which one the model generalizes

[Joshua Ashkinaze @JoshAshkinaze]

@postingcontent I believe this is just a todo list + maintenance of said list. The problem is apps only do the first part. The Notion feature team needs to add mind-reading.

[Jürgen Pfeffer @jurgenpfeffer.bsky.social @JurgenPfeffer]

Congratulations to @MiriamSchirmer for successfully defending her Ph.D. „Natural Language Processing for Violence Studies: Investigating Trauma and Online Aggression“ - thanks to the committee @fredmorstatter and Sarah Diefenbach.

[ACM CSCW @ACM_CSCW]

How does placing users in different POVs in VR-simulated moral dilemmas affect people’s decision-making? In a new #CSCW2024 paper, @yinsecurity1 considers this question and its implications for building meaningful and empathy-inducing VR experiences: https://t.co/tca8kXQUGo

[Joshua Ashkinaze @JoshAshkinaze]

@manoelribeiro Interesting writeup! If changes to the algorithm are things Meta *would have done anyway* (absent experiment), it makes conclusions harder to interpret but not fully undermines the study---goal is to compare "base case" (which may involve updates etc) against chronological feed.

[Ethan Mollick @emollick]

This paper is a really nice example of using LLMs as a co-intelligence for high-end research, even when the AI can't really do the work itself, by helping explore an idea space & surfacing unexpected directions

The value of a creative partner, even in quantitative work, is high

[Chase @ChaseMc67]

Someone just leaked the @OpenAI text watermark code

[WikiResearch @WikiResearch]

Study participants perceived the same text as *less* credible when presented as a Wikipedia article than when encountering it as simulated ChatGPT or Alexa output

[School of Information @umsi]

Meet UMSI alumna and @Jeopardy super-champion Adriana Harmeyer (MSI ’12)! Harmeyer, an archivist at Purdue University, takes us behind the scenes of her 15-game winning streak in this Q&A: https://t.co/ROHBLrJBAY
@LifeAtPurdue

[Matt Groh @mattgroh]

On my way to @IC2S2 and psyched to tune into the awesome set of talks and posters!

At the poster session tomorrow, @n3gRain is presenting our latest work around how to build generative AI literacy.

If you're around, say hi!

[IC2S2 @IC2S2]

Thanks EVERYONE for a great #ic2s2 this year! Take a few minutes filling out the feedback survey and help us improve: https://t.co/INJhqOGKrL

[Allie K. Miller @alliekmiller]

A new study by the Max Planck Institute and others reveals that lie-detection algorithms outperform humans.

Participants were more likely to accuse someone of lying when AI supported their decision, especially when they actively sought the algorithm's advice.

- Algorithm accuracy: 66.86% vs. human 46.47%

- Accusation rate: 19.22% (no AI) to 30.39% (with AI)

- Algorithm influence: 31% requested AI help (most followed its advice, increasing accuracy of accusations)

Read more here: https://t.co/pFV87G5Pnk

[WikiResearch @WikiResearch]

In the latest issue of our monthly newsletter:
• "Is Wikipedia Politically Biased?" Perhaps

And other recent publications, about low-resource languauges and LLMs, GLAM bias, Wikipedia vs. academic self-promotion, and more
https://t.co/yzGVUb50CA

[WikiResearch @WikiResearch]

"Seeing Like an AI: How LLMs Apply (and Misapply) Wikipedia Neutrality Norms" Are LLMs’ capable to detect and correct biased Wikipedia edits according to Wikipedia��s Neutral Point of View (NPOV) policy?

(@JoshAshkinaze et al, 2024)

https://t.co/lzz455gb0i
@eegilbert

[School of Information @umsi]

New research from UMSI:
Prompt Design Matters for Computational Social Science Tasks but in Unpredictable Ways
Shubham Atreja, Joshua Ashkinaze, Lingyao Li, Julia Mendelsohn, Libby Hemphill
@ShubhamAtreja @JoshAshkinaze @jmendelsohn2 @libbyh @arxiv
https://t.co/7qUBMLGLXI

[Ethan Mollick @emollick]

It gets better: Grand Theft Auto games lower crime rates

When a GTA game is released, violence & crime actually drop, as potential criminals stay inside to play (This applies to other big shooters, too, so the yearly CoD releases are actually good for the world, sorry to say).

[Joshua Ashkinaze @JoshAshkinaze]

@WikiResearch Thanks for sharing!

[WikiResearch @WikiResearch]

LLMs such as ChatGPT struggle with identifying neutrality violations on Wikipedia ("better than random individuals...but worse than expert editors")
But "crowdworkers rated AI rewrites as more neutral (70%) and fluent (61%) than Wikipedia-editor rewrites" https://t.co/mPKcMeSv3b

[MichiganAI @michigan_AI]

Get ready for an event like no other! 🚀
🎉 Excited to announce that our upcoming #FridayNightAI Q&A event will include interactive #AI activities by @JoshAshkinaze and Zilin Wang!

Save the date and get ready to be part of an unforgettable experience!

https://t.co/sQRtd6uO2A

[Hause Lin @hauselin]

Updated the extension with these new features (thanks everyone for suggestions!)
- works for quote tweets
- more customization (e.g., on/off toggle, add your own URLs/accounts, enable post highlight)

https://t.co/LcdeHBcucH