as AI increasingly supports shopping and ads, it’s worth remembering that retrieval often shapes who gets exposure in final generated output. in a recent paper, @teknology.bsky.social uses methods from fair ranking to assess and address exposure bias in downstream generation.
841.io/doc/fairrag....
Explainable AI is often assumed to build trust. A study of sonographers estimating gestational age found AI predictions improved accuracy, but explanations did not. In fact, explanations made some clinicians perform worse, highlighting user variability.
#MedSky #MLSky
Thrilled to be selected for the 🎓 Rising Stars in Data Science Workshop! Grateful to @stanforddata.bsky.social, @HCID UC San Diego, and @dsi-uchicago.bsky.social for this opportunity.
Excited to share my work on trustworthy and collaborative AI and connect with amazing peers and mentors.
🔗 👇
Yes, more so with code for running quick experiments! i definitely want my code to NOT fail gracefully. (And save myself hours of debugging time because there is a default parameter somewhere I did not notice!)
Ah that makes sense! Thanks, yeah I am on that slack, hhh!
How can I get an invite for the XAI discord?
Thank you for making the list, could you please add me?
In a stunning moment of self-delusion, the Wall Street Journal headline writers admitted that they don't know how LLM chatbots work.
What if you could understand and control an LLM by studying its *smaller* sibling?
Our new paper introduces the Linear Representation Transferability Hypothesis. We find that the internal representations of different-sized models can be translated into one another using a simple linear(affine) map.
McCombs article: news.mccombs.utexas.edu/research/to-...
Paper url: doi.org/10.47989/ir3...
@utaustin.bsky.social
@texasscience.bsky.social
@engagingnews.bsky.social
@utischool.bsky.social
#TexasAI
#YearofAI
Can content moderation models balance accuracy & fairness?
UT McCombs news featured our iConference paper by Soumyajit Gupta on optimizing the fairness-accuracy tradeoff in toxicity detection. In collaboration with Venelin Kovatchev @mariadearteaga.bsky.social @mattlease.bsky.social
How good are LLMs at 🔭 scientific computing and visualization 🔭?
AstroVisBench tests how well LLMs implement scientific workflows in astronomy and visualize results.
SOTA models like Gemini 2.5 Pro & Claude 4 Opus only match ground truth scientific utility 16% of the time. 🧵
Please join us for the TrustNLP workshop (215 San Miguel) @naaclmeeting.bsky.social #trustNLP2025
Session detail:
Poster Session 5 - IAM: Interpretability and Analysis of Models for NLP, Hall 3
This is a collaborative work with Manoj Kumar, Ninareh Mehrabi, Anil Ramakrishna, Anna Rumshisky, Kai-Wei Chang, Aram Galstyan, Morteza Ziyadi, Rahul Gupta
Causal tracing informed edits provide a better detoxification-degeneration trade-off.
Model editing helps reduce toxicity. High detoxification can be achieved by simply editing random MLP layers. However, this leads to degeneration and increased perplexity.
We find evidence of toxic memory in the early layer of GPT-2 XL for innocuous-looking adversarial prompts.
Paper: On Localizing and Deleting Toxic Memories in Large Language Models
Anthology URL: aclanthology.org/2025.finding...
Excited to present my internship work at
Amazon AGI at @naaclmeeting.bsky.social tomorrow at 2:00 pm local time. Please come say hi if you are around.
thinking of calling this "The Illusion Illusion"
(more examples below)
Created a small starter pack including folks whose work I believe contributes to more rigorous and grounded AI research -- I'll grow this slowly and likely move it to a list at some point :) go.bsky.app/P86UbQw
NeurIPS Test of Time Awards:
Generative Adversarial Nets
Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio
Sequence to Sequence Learning with Neural Networks
Ilya Sutskever, Oriol Vinyals, Quoc V. Le
Right, sorry for being unclear. I saw your comment sharing the Qualtrics integration tutorial with a video. bsky.app/profile/dggo...
Nvm, found it!
@tomcostello.bsky.social 's Qualitrics materials and tutorial video for integrating LLMs into Qualtrics can be accessed at publish.obsidian.md/qualtrics-do...
Will there be a video for this talk?
🙋🏽
