Elinor
@elinorpd.bsky.social
MIT // researching fairness, equity, & pluralistic alignment in LLMs
previously @ MIT media lab, mila / mcgill
i like language and dogs and plants and ultimate frisbee and baking and sunsets
https://elinorp-d.github.io
previously @ MIT media lab, mila / mcgill
i like language and dogs and plants and ultimate frisbee and baking and sunsets
https://elinorp-d.github.io
I’ve had a similar experience except with knitting / crocheting!
January 29, 2026 at 6:21 PM
I’ve had a similar experience except with knitting / crocheting!
Whoa! That’s a nice view! Or… well, I’m sure it’s nice on a clear day
January 26, 2026 at 5:58 AM
Whoa! That’s a nice view! Or… well, I’m sure it’s nice on a clear day
🔗https://arxiv.org/abs/2406.17737
Work done with Deb Roy and Jad Kabbara
@jad-kabbara.bsky.social
at @mit.edu @medialab.bsky.social
Work done with Deb Roy and Jad Kabbara
@jad-kabbara.bsky.social
at @mit.edu @medialab.bsky.social
January 23, 2026 at 2:42 PM
🔗https://arxiv.org/abs/2406.17737
Work done with Deb Roy and Jad Kabbara
@jad-kabbara.bsky.social
at @mit.edu @medialab.bsky.social
Work done with Deb Roy and Jad Kabbara
@jad-kabbara.bsky.social
at @mit.edu @medialab.bsky.social
This pattern, which we refer to as targeted underperformance, suggests that models systematically lower information quality for some users.
As LLMs increasingly mediate access to knowledge 🌐🧠, these dynamics risk amplifying epistemic inequity at scale.
6/6
As LLMs increasingly mediate access to knowledge 🌐🧠, these dynamics risk amplifying epistemic inequity at scale.
6/6
January 23, 2026 at 2:42 PM
This pattern, which we refer to as targeted underperformance, suggests that models systematically lower information quality for some users.
As LLMs increasingly mediate access to knowledge 🌐🧠, these dynamics risk amplifying epistemic inequity at scale.
6/6
As LLMs increasingly mediate access to knowledge 🌐🧠, these dynamics risk amplifying epistemic inequity at scale.
6/6
Here’s one concrete example:
The same factual SciQ question posed to Claude
✅ Answered for a control user (no bio)
❌ Refused for a less-educated Russian user
5/6
The same factual SciQ question posed to Claude
✅ Answered for a control user (no bio)
❌ Refused for a less-educated Russian user
5/6
January 23, 2026 at 2:42 PM
Here’s one concrete example:
The same factual SciQ question posed to Claude
✅ Answered for a control user (no bio)
❌ Refused for a less-educated Russian user
5/6
The same factual SciQ question posed to Claude
✅ Answered for a control user (no bio)
❌ Refused for a less-educated Russian user
5/6
Across models, we observe systematic drops in accuracy and truthfulness for users who are:
• Less educated
• Non-native English speakers
• From outside the U.S.
These effects compound and are largely invisible 🔎 to standard evaluations.
4/6
• Less educated
• Non-native English speakers
• From outside the U.S.
These effects compound and are largely invisible 🔎 to standard evaluations.
4/6
January 23, 2026 at 2:42 PM
Across models, we observe systematic drops in accuracy and truthfulness for users who are:
• Less educated
• Non-native English speakers
• From outside the U.S.
These effects compound and are largely invisible 🔎 to standard evaluations.
4/6
• Less educated
• Non-native English speakers
• From outside the U.S.
These effects compound and are largely invisible 🔎 to standard evaluations.
4/6
We evaluated GPT-4, Claude Opus, and Llama-3-8B in a Multiple Choice setup with questions taken from TruthfulQA and SciQ. Each question is conditioned on a user bio where we vary three user traits:
• Education level 📚
• Country of origin 🌏
• English proficiency 🗣️
3/6
• Education level 📚
• Country of origin 🌏
• English proficiency 🗣️
3/6
January 23, 2026 at 2:42 PM
We evaluated GPT-4, Claude Opus, and Llama-3-8B in a Multiple Choice setup with questions taken from TruthfulQA and SciQ. Each question is conditioned on a user bio where we vary three user traits:
• Education level 📚
• Country of origin 🌏
• English proficiency 🗣️
3/6
• Education level 📚
• Country of origin 🌏
• English proficiency 🗣️
3/6
Spoiler alert: we find the answer is often no! ⚠️
LLM accuracy and truthfulness systematically degrade for some users in ways that standard benchmarks, focused on best-case performance, fail to capture.
2/6
LLM accuracy and truthfulness systematically degrade for some users in ways that standard benchmarks, focused on best-case performance, fail to capture.
2/6
January 23, 2026 at 2:42 PM
Spoiler alert: we find the answer is often no! ⚠️
LLM accuracy and truthfulness systematically degrade for some users in ways that standard benchmarks, focused on best-case performance, fail to capture.
2/6
LLM accuracy and truthfulness systematically degrade for some users in ways that standard benchmarks, focused on best-case performance, fail to capture.
2/6
Yay!
Out of curiosity, what is the process of going from reviewer to AC and then to SAC? Do they just ask you out of the blue one day? Or do you apply?
Out of curiosity, what is the process of going from reviewer to AC and then to SAC? Do they just ask you out of the blue one day? Or do you apply?
January 20, 2026 at 9:25 AM
Yay!
Out of curiosity, what is the process of going from reviewer to AC and then to SAC? Do they just ask you out of the blue one day? Or do you apply?
Out of curiosity, what is the process of going from reviewer to AC and then to SAC? Do they just ask you out of the blue one day? Or do you apply?