Log inSign up
Image
user avatar
Vishakh Padmakumar
@vishakh_pk
Measuring and mitigating the societal impacts of AI/LLMs @stanfordnlp @stanfordAILab Prev @allen_ai @NYUDataScience
vishakhpk.github.io
Joined August 2015
693
Following
864
Followers
RepliesMedia
  • Pinned
    user avatar
    Vishakh Padmakumar
    @vishakh_pk
    Jun 3
    People are increasingly worried that AI tools make us overreliant. But how do we actually measure this? We introduce Offloading Score, a measure of reliance based on the fraction of cognitive effort offloaded to AI while completing a task. In a controlled user study, Offloading
    Image
    Image
  • user avatar
    Vishakh Padmakumar
    @vishakh_pk
    Jun 8
    It's been crazy to see the effects of content homogenization play out in real time over the past few years, even as contemporary models get stronger and stronger! Such important work from @YekyungKim @YapeiChang @MohitIyyer, and a nudge to our fears from 2023 -
    user avatar
    Yekyung Kim
    @YekyungKim
    Jun 8
    From op-eds in newspapers to NeurIPS position papers, AI is increasingly shaping long-form public discourse. Its arguments seem plausible, but beneath surface fluency, we find argument collapse: different LLMs converge to the same main & supporting arguments and structure.
    Image
  • user avatar
    Vishakh Padmakumar
    @vishakh_pk
    Jun 3
    Thanks again for all your guidance on this project! As our AI tools and interaction modes evolve, we also need new ways to evaluate their impact and make sure they continue to support meaningful user outcomes.
    user avatar
    Vera Liao
    @QVeraLiao
    Jun 3
    Overreliance is the root of many problems with AI use, from errors to deskilling and loss of human agency. But with GenAI and now agentic tools, we have to rethink how to conceptualize and measure AI reliance. Really excited to see this work out!
  • user avatar
    Vishakh Padmakumar
    @vishakh_pk
    Apr 29
    During my thesis defense, @eunsolc asked me a question about how to best ensemble the most diverse set of responses for a prompt given a suite of LLMs. I didn't know, but @YuhanLiu_nlp answers expertly through this work!!πŸ§΅πŸ’‘
    user avatar
    Yuhan Liu
    @YuhanLiu_nlp
    Apr 28
    Can LLMs generate diverse outputs for open-ended questions? Is it helpful if we ensemble outputs from multiple models? We study 18 LLMs on 4 datasets and find that no single model is best at generating diverse outputs πŸ‘‡/ 🧡
    Image