Elisa Kreiss

The normalization of (almost) everything: Our minds can get used to anything, and even crises start feeling normal Our minds can get used to anything, and even crises start feeling normal

Honored and excited to share that I am the winner of Nomis & Science Young Explorer Award!!

Also thrilled to share that my article describing my research is out now in @science.org today!

The normalization of (almost) everything www.science.org/doi/10.1126/...

1/

Now that the ICLR deadline is behind us, happy to share that From Language to Cognition has been accepted as an Oral at #EMNLP2025! 🎉

Looking forward to seeing many of you in Suzhou 🇨🇳

For folks considering grad school in ML, my advice is to explore programs that mix ML with a domain interest. ML programs are wildly oversubscribed while a lot of the fun right now is in figuring out what you can do with it

Keynote spotlight #3: COLM's second day will open with some *real* intelligence 🧠🧠🧠 with Tom Griffiths from Princeton

I finally got around to making a tool to compare completions from SFT vs. RLHF trained models. This is a mini site for the RLHF book that I've wanted for a while.

buff.ly/lqDL5wa

It's always been hard to say what RLHF does to a model within a more complex post-training pipeline.

Keynote spotlight #1:
COLM will open on Oct 7 with a keynote by
@lukezettlemoyer.bsky.social about mixed-modals models 🌈

It is again the time of year when I beg @aclmeeting.bsky.social execs to rethink the current streaming platform system. For my #EMNLP2025 submissions, I am *required* to upload 2 video recordings + 2 posters + 2 slide decks. Why force both posters and talks for all? Nonsense.

New paper! We find that the more prompts "look like" common benchmark tasks, the more model behavior changes. Even when we encourage the model to explicitly infer gender, the instruction itself causes the opposite, namely boosting gender-neutral language. Check out Bufan's thread and our paper!

Violin plots of Probability of Pronoun Shift. Models show significant sensitivity to prompt changes: when prompts highlight gender evaluation, pronoun use shifts, with decreased “he” and increased “they” use.

🚨 New #EMNLP2025 paper!

Do LLMs exhibit distinct behavior when the prompt looks similar to common evaluation prompts? 👀

We show that prompts that signal bias evaluation can flip the measured bias. See below ⬇️

Department of Communication - Mediated Interpersonal Communication University of California, Davis is hiring. Apply now!

We’re hiring @ucdavis.bsky.social Comm! TT Asst Prof in Mediated Interpersonal Communication. Our dept blends behavioral, biological & computational strengths at one of the most interdisciplinary campuses anywhere. Apps due Oct 15. Glad to chat recruit.ucdavis.edu/JPF07289

My lab at UCLA is hiring 1-2 PhD students this cycle!

Join us to work at the intersection of cognitive science and AI applied to pressing societal challenges like climate change.

More info about me: rachit-dubey.github.io

My lab: ucla-cocopol.github.io

Please help repost/spread the word!

How can an imitative model like an LLM outperform the experts it is trained on? Our new COLM paper outlines three types of transcendence and shows that each one relies on a different aspect of data diversity. arxiv.org/abs/2508.17669

The 2025-2026 #academic job market season is upon us.

Here are some resources and #job postings for #faculty and #postdoc positions for those in Information/Computer/Social Sciences, and related fields.

Will update the list as more #jobs are posted.

🧵(0/n)

#AcademicChatter

New paper! We explore a radical paradigm for AI evals: assessing LLMs on *unsolved* questions.

Instead of artificially difficult exams where progress ≠ value, we assess LLMs on organic, unsolved problems via reference-free LLM validation & community verification. LLMs solved ~10/500 so far:

Playing telephone with generative models: "verification disability," "compelled reliance," and accessibility in data visualization This paper is a collaborative piece between two worlds of expertise in the field of data visualization: accessibility and bias. In particular, the rise of generative models playing a role in accessibi...

Preprint announcement! 🎉

I've written a short position paper with Cindy Xiong Bearfield (a bias researcher in visualization) for @ieeevis.bsky.social Workshop on Accessibility on how using LLMs and generative models forces reliance (rather than builds trust) when used for accessibility.

UMich Ling is hiring for a cluster hire position in AI and Linguistics! Come work in a fabulous department with great grad students, supportive colleagues, and one overly school-spirited nut who will tell you where all the fun stuff is on campus.

apply.interfolio.com/171006

This is the first example I've seen of a university trying to mobilize constituents to stand up against Trump threats.

There's been a lot of "wHaT eLsE CaN AdMiNiStRaToRs Do?" in The Discourse when this is exactly the thing they haven't been doing. Good for UCLA for trying.

Others should follow.

Cartoon of a scientist peering through a microscope saying "Whoaah!!!". Title says "The best part of Science is knowing, for a moment, something that nobody else in the world knows."

Really just had something like this feeling this afternoon. It’s the high that keeps you coming back for more

Excellent non-technical explanation! Let’s raise AI literacy without scaring people off with overly complicated terminology while preserving technical accuracy!

EPFL NLP Postdoctoral Scholar Posting - Swiss AI LLMs The EPFL Natural Language Processing (NLP) lab is looking to hire a postdoctoral researcher candidate in the area of multilingual LLM design, training, and evaluation. This postdoctoral position is as...

The EPFL NLP lab is looking to hire a postdoctoral researcher on the topic of designing, training, and evaluating multilingual LLMs:

docs.google.com/document/d/1...

Come join our dynamic group in beautiful Lausanne!

Neat year #NLP conferences:
📍 #EACL2026: 24–29 March 🇲🇦 Rabat (ARR deadline: 6 Oct)
📍 #LREC2026: 11–16 May 🇪🇸 Palma de Mallorca (deadline: 17 Oct)
📍 #ACL2026: 2–7 July 🇺🇸 San Diego (ARR deadline: 6 Jan)
📍 #COLING2027: 9–14 May 2027 🇲🇴 Macao

Only 2 out of 120 papers in my NeurIPS SAC batch have all reviewer scores ≥4! Don't despair with the low scores and focus on writing a clear and concise rebuttal! Good luck!

‘Boiling frog’ effect makes people oblivious to threat of climate crisis, shows study Research shows people’s response to binary data could drive more effective ways of communicating gravity of crisis

Woke up to a pleasant surprise! Guardian wrote a nice article covering our Binary climate data study!

TLDR: “Before and after” climate data can help counter the boiling frog effect

www.theguardian.com/news/2025/ju...

Headed to #CogSci2025 next week!

Interested in computational cognitive science applied to climate change and real-world policy? Come find me!

My lab at UCLA is recruiting 1–2 fully funded PhD students.

We work on:
• decision-making
• climate perception
• human–AI collaboration for policy design

For Prospective PhD Students I’m recruiting PhD students who will begin their degree in Fall 2026! I am an incoming assistant professor at Wisconsin-Madison’s Computer Sciences department, and my research focuses on natural langu...

I'm sadly not at #IC2S2 😭, but I will be at #ACL2025 in Vienna ☕️ next week!!

Please spread the word that I'm recruiting prospective PhD students: lucy3.notion.site/for-prospect...

We're hiring!

academicjobsonline.org/ajo/jobs/30219

The Cornell Department of Communication is hiring an Assistant Professor in Science, Environment, and Risk Communication!

Applications received by September 15, 2025, will be given full consideration.

You have a budget to human-evaluate 100 inputs to your models, but your dataset is 10,000 inputs. Do not just pick 100 randomly!🙅

We can do better. "How to Select Datapoints for Efficient Human Evaluation of NLG Models?" shows how.🕵️
(random is still a devilishly good baseline)

The mystery of em‑dashes: part two with quantitative evidence A couple of weeks ago I made an assumption: the rise of em‑dashes in AI‑generated text happened because model providers started scanning older, pre‑Kindle books.

This is the best explanation I've seen yet for _why_ language models prefer em-dashes (—). In brief: The tokenization scheme mean em-dashes result in a smaller loss than other, equivalent punctuation options.
msukhareva.substack.com/p/the-myster...

Experimentology cover: title and curves for distributions.

Experimentology is out today!!! A group of us wrote a free online textbook for experimental methods, available at experimentology.io - the idea was to integrate open science into all aspects of the experimental workflow from planning to design, analysis, and writing.

Can you train a performant language model using only openly licensed text?

We are thrilled to announce the Common Pile v0.1, an 8TB dataset of openly licensed and public domain text. We train 7B models for 1T and 2T tokens and match the performance similar models like LLaMA 1 & 2