Eytan Adar

Eytan Adar

@eytan.adar.prof

Michigan faculty, http://www.cond.org

1.7K followers2.4K following385 posts

Eytan Adar
@eytan.adar.prof

I grew up there, so find "Cupertino" amusing (mostly lawyer drama, but some skewering of SV). In the latest episode one of the characters searches for "people who hate silicon valley." The list is *mostly* fake, but at the top are some familiar faces.

Eytan Adar
@eytan.adar.prof

We are running a UIST workshop on revibing--the idea that technical HCI papers can be reconstructed from the paper itself. We're interested in the implications for evaluation, research, and education. Join us! revibe-hci.github.io

Eytan Adar
@eytan.adar.prof

I wrote a post about the emergence of AI-native students who do their research using GenAI tools. I think that there are some real risks to job prospects and the likely changes in their fields.

Eytan Adar
@eytan.adar.prof

Some of the same things that make these students fast today will cost them soon. There are some perverse incentives where we (faculty) are not addressing this.

Eytan Adar
@eytan.adar.prof

Before: Claude feeds human ADHD. Now: Claude gets ADHD... "You're right — I got pulled into the distractor-similarity bug report and never actually shipped the color feature, I only did the palette research. Let me finish it now." (after I told it that it didn't finish what I asked for)

Eytan Adar
@eytan.adar.prof

if elections only reported the results once all the counts were in, what would the "elections are rigged" people complain about next?

Eytan Adar
@eytan.adar.prof

Maybe obvious, but working theory: The number of times a paper has been rejected is ~the difference between the latest model version and the llm model used in the paper

Eytan Adar
@eytan.adar.prof

Inspired by this thread: bsky.app/profile/tedu... (but it's not about taxing the LLM vendors, but the users... Chevron/OpenAI can pay, but so can the people who fill up/use the resource)

Ted Underwood@tedunderwood.com

Good post by a history prof, in the genre “What most people don’t understand yet is …” — although, in my experience, a lot of the academics it’s addressing do grasp what’s happening with agents, and just really wish it would not happen.

generativehistory.substack.comThe Agents Are Waking UpThe Intelligence Revolution that swept through the software industry this past winter is coming to knowledge-work next.
Eytan Adar
@eytan.adar.prof

Almost a year to the day since I last posted this. An admonishment by Mao's government to the China's statisticians on the need for "statistics with particular Party character." (Chinese newspaper in '61)

southpaw@nycsouthpaw.bsky.social

The gallup presidential approval poll is an institution in American politics that predates world war 2. With Trump’s approval in the 30s, they’re ending it. thehill.com/homenews/med...

Eytan Adar
@eytan.adar.prof

Can someone explain to me the objections to the arxiv survey paper thing? I get that it might be a biased, and maybe not grounded in data, but what's the actual/perceived benefit for posting on arxiv vs... I don't know... anywhere else? Is it citations? visibility? genuinely curious...

Eytan Adar
@eytan.adar.prof

FWIW, the studies I've seen on benefits of prepubs for citations is a mixed bag. But these seem to focus on published vs published + arxiv preprint. Maybe it's different for never published + preprint on arxiv vs preprint elsewhere

Eytan Adar
@eytan.adar.prof

Coining a new phrase: "Great Pacific Paper Patch." Definition: The growing collection of rejected papers floating from conference to conference and clogging up the knowledge ocean. :)

Eytan Adar
@eytan.adar.prof

When reacting to some kind of new (manufactured) scandal every day, the press forgets they can come back the next day and say: you didn't know about it yesterday, but you've had a day to think about this, what's your reaction?

Aaron Rupar@atrupar.com

Q: What's your reaction to yesterday's purge of senior ICE officials? When is it time for Congress to do some oversight? MIKE JOHNSON: I don't know any of the details of that yet. I just heard about that, literally I was walking in.

Eytan Adar
@eytan.adar.prof

Fun paper from a recent project with UMich alumni Tenghao Ji. Can we pick better images for Wikipedia articles? Given all the choices in the Wikipedia Commons, which image is best as an "instructional aid?" (1/5)

View full thread
Eytan Adar
@eytan.adar.prof

Images selected in this way are better representatives of the main properties of the concept, while also highlighting what makes it different from other, related concepts. The paper with more details is here: arxiv.org/abs/2509.15059 (5/5)

arxiv.orgQuizRank: Picking Images by Quizzing VLMsImages play a vital role in improving the readability and comprehension of Wikipedia articles by serving as `illustrative aids.' However, not all images are equally effective and not all Wikipedia edi...
Eytan Adar
@eytan.adar.prof

I should also add that you don't need the LLM/VLM to do this... you can have humans do some (or all) of the algorithm. It's just that the LLM/VLM solution makes it scale and can highlight images for editors that they might not have considered. (6/5)

Eytan Adar
@eytan.adar.prof

Fun paper from a recent project with UMich alumni Tenghao Ji. Can we pick better images for Wikipedia articles? Given all the choices in the Wikipedia Commons, which image is best as an "instructional aid?" (1/5)

View full thread
Eytan Adar
@eytan.adar.prof

So if I ask: "what is the chest color of the Western Bluebird?", a good image helps you answer correctly (orange). QuizRank uses an LLM to generate questions (based on the article) and a VLM to take the test. Images that help the VLM do well on the test are better and are ranked more highly. (3/5)

Eytan Adar
@eytan.adar.prof

We also extended this idea to build a Contrastive QuizRank. Instead of just considering what is interesting about the concept (e.g., the Western Bluebird), we let the LLM figure out what makes it different from a "distractor" (e.g., the Mountain Bluebird). (4/5)

Eytan Adar
@eytan.adar.prof

Fun paper from a recent project with UMich alumni Tenghao Ji. Can we pick better images for Wikipedia articles? Given all the choices in the Wikipedia Commons, which image is best as an "instructional aid?" (1/5)

Eytan Adar
@eytan.adar.prof

For example, given the 234 different images of the Western Bluebird, which should I pick? Our intuition is that with a "good" instructional image, someone should be able to answer questions about important visual properties of the concept better than with a "bad" image. (2/5)

Eytan Adar
@eytan.adar.prof

We need something more technical sounding than "co-evolution" to describe this... RHLAF: reinforcing human learning through AI feedback? HAAF: human adaptation through AI feedback? :)

mr. TIM@timkellogg.me

As LLMs Improve, People Adapt Their Prompts a study shows that a lot of the real world performance gains that people see are actually because people learn how to use the model better arxiv.org/abs/2407.14333

Eytan Adar
@eytan.adar.prof

Who knew... Claude can be funny... ;) Real researchers don't use LLMs: claude.ai/share/91111e...

claude.aiClaudeTalk with Claude, an AI assistant from Anthropic
Eytan Adar@eytan.adar.prof

Taken to the extreme this starts to sound like real programmers don't eat quiche (www.bernstein-plus-sons.com/RPDEQ.html). Real researchers don't use LLMs. Real researchers don't use search engines. Real researchers don't use anything other than articles faxed to them from the author.

bernstein-plus-sons.comReal Programmers Don't Eat Quiche

Eytan Adar reposted

Jessica Hullman
@jessicahullman.bsky.social

Decision-making studies appear in HCI, vis, & increasingly AI/ML, but how “good decision” is defined is often ad-hoc My #CHI2025 talk today will answer Qs like: What's a decision problem? What's the best possible performance on a decision problem? What minimum info must participants be given? 1/2

Eytan Adar
@eytan.adar.prof

We need to stop talking about indirects as: oh, I need $100k for research, but we charge the government $155k so I can give $55k to my institution for whatever. 1/2

Eytan Adar
@eytan.adar.prof

It should be: I need $155k to do my research. Some of that goes towards paying grad students, lab equipment, etc., and some of it goes to pay for the building, admin help, power, IT services, etc. Which is actually what is going on 2/2

Eytan Adar
@eytan.adar.prof

The new administration is just crazy. What other organization achieves the same policy effect no matter who gets hired? They can bring in the super strategic or super incompetent and the results (at least short term) are the same.

Eytan Adar
@eytan.adar.prof

Consider a fictional graduate student brain N, and a scientist Q who wishes to use N in their research pipeline (e.g. for analysis and/or interpretation). Functionally, N is not fully understood by Q. Is it acceptable to use N? To be clear, there's a difference, but not sure it's "understanding."

Iris van Rooij 💭@irisvanrooij.bsky.social

Scientific Integrity Question Consider a fictional AI system M, and a scientist S who wishes to use M in their research pipeline (e.g., for analysis and/or interpretation). Functionally, M is not fully understood by S. In your opinion, is it acceptable that S uses M in their research pipeline?

Eytan Adar
@eytan.adar.prof

The regulation of AI debate feels like that scene out of Contact where the government panel tells Jodie Foster's character that she can't go meet the aliens because she doesn't represent humanity. But then some crazy billionaire is like, whatevs...