mariannealq.bsky.social
Information Science PhD @ Cornell Tech. Looking at AI's impact on information ecosystems & news consumption. mariannealq.com
53 posts
1,377 followers
293 following
Prolific Poster
Conversation Starter
comment in response to
post
Boring answer but I suspect this is almost entirely down to long publishing cycles. Papers that were published Nov 2024 were mostly submitted Jan 2023, and ChatGPT was launched Nov 2022. Compared with CHI2024, where papers were submitted Sept 2023
comment in response to
post
So exciting!!! Congrats 🥳
comment in response to
post
Not sure if 100% what you’re looking for but I have a paper applying an inductive qual codebook using an LLM: dl.acm.org/doi/abs/10.1... & @hopeschroeder.bsky.social and I just got our newest paper on LLMs for qual research accepted to CHI (posting the CR sometime this week) arxiv.org/abs/2410.07362
comment in response to
post
agree — as a euro expat, I think the “American researchers will flock to Europe” narrative is rooted in an oversimplified conception of what European countries stand for; & doesn’t take into account idiosyncracies of research in other countries or Europe’s own move to the far right the last years.
comment in response to
post
Thanks for joining us at the workshop!
View all the accepted papers here: sites.google.com/view/llmsind...
@hopeschroeder.bsky.social @caseyrandazzo.bsky.social
comment in response to
post
We conclude with (paraphrased) words from @dmimno.bsky.social about why HCI needs to be at the center of these conversations: “The underlying technology may not be what’s going to cause a revolution; user interaction systems that make them easy to use will.”
comment in response to
post
2) The field desperately needs better standards to evaluate, validate, and replicate LLM outputs.
3) Researchers, especially junior scholars, need developed and accepted norms to use these tools with confidence and without fear of backlash.
comment in response to
post
1) LLMs are permeating the HCI research pipeline. As has now been empirically shown (e.g. arxiv.org/abs/2411.05025, arxiv.org/abs/2410.07362), these tools are being used broadly and touch many different facets of research.
comment in response to
post
Thanks Raj, it so rarely works out that way!
comment in response to
post
haha, I would think that's pretty easy to do!
comment in response to
post
By modeling average concreteness, we can show that when all headlines are generally vague, increasing concreteness predicts more clicks. When all headlines are more concrete, increasing concreteness eventually predicts fewer clicks.
comment in response to
post
Thanks Jeremy! Good question — we included average concreteness in an experiment as an interaction term to test whether headline concreteness has different effects at high vs low levels of concreteness for the same article. As you said, this helps control topic, date, different images, etc.
comment in response to
post
7/ Excited to share this work and make the concreteness measure publicly available for other researchers!
Study: www.nature.com/articles/s41...
Website: concreteness.mariannealq.com
Concreteness measure: github.com/maubinle/sen...
comment in response to
post
6/ Headlines that strike a balance—revealing just enough to spark curiosity—get the most clicks. This suggests a "Goldilocks effect" where revealing the right amount of information maximizes engagement.
These findings also harmonize prior conflicting findings, and tie back to the information gap.
comment in response to
post
5/ To test this theory in a news context, we tagged nearly 9,000 real headline experiments with a novel automated concreteness measure that labels how much specific information a headline reveals. As it turns out, both vague and overly detailed headlines perform worse.
comment in response to
post
4/ So, we went back to the original concept of the “information gap” from the psychology of curiosity. Information gap theories suggests that people are most curious in medium-information environments: when some information, but not all, is available.
comment in response to
post
3/ The scientific evidence was split – some studies found that using linguistic curiosity features like using pronouns or numbers in headlines could increase clicks, while others actually found negative effects.
comment in response to
post
2/ Headline writers regularly bank on “curiosity gap” headlines to drive engagement: headlines that purposefully obscure information to create a sense of mystery and incite more clicks. But do these actually work?