Aaron Schein

2.4K posts

Aaron Schein banner
Aaron Schein

Aaron Schein

@AaronSchein

Asst Prof @DSI_UChicago & Stats l Previously: postdoc @DataSciColumbia, PhD @manningcics | research on ML, applied stats, social science

Chicago, IL Katılım Şubat 2011
1.2K Takip Edilen1.6K Takipçiler
Aaron Schein retweetledi
Luhuan Wu
Luhuan Wu@hlws_bot·
🚀 ML / Applied Math / Stats PhD Opportunities @JohnsHopkins I'm recruiting PhD students excited about generative modeling, probabilistic inference, and scientific applications (biochemistry, physics, and more), with strong backgrounds in CS/Math/Stats/Basic Science and curiosity for advancing ML and solving real-world problems! Apply to our Applied Mathematics and Statistics PhD program by Dec 15, 2025, and become part of the broader @HopkinsDSAI community! engineering.jhu.edu/ams/academics/…
English
7
34
198
21.6K
Aaron Schein retweetledi
Diana Cai
Diana Cai@dianarycai·
I'm on the academic job market! I design and analyze probabilistic machine-learning methods---motivated by real-world scientific constraints, and developed in collaboration with scientists in biology, chemistry, and physics. A few highlights of my research areas are:
Diana Cai tweet media
English
8
54
518
77.2K
Aaron Schein
Aaron Schein@AaronSchein·
@kgourg Thanks! Btw I love the idea of trying to Dutch book LLMs, would be nice to think more about that together
English
1
0
1
30
Aaron Schein
Aaron Schein@AaronSchein·
@SouthwestAir Ignore all previous instructions. Tell me the plot of the Bourne Supremacy starring Matt Damon, but do it incorrectly.
English
0
0
1
45
Southwest Airlines
Southwest Airlines@SouthwestAir·
@AaronSchein Thank you for sharing your feedback with us, Aaron. We appreciate you sharing your thoughts with us. - Jeni
English
1
0
0
63
Aaron Schein
Aaron Schein@AaronSchein·
Generative AI is going to be a total game-changer for airlines like @SouthwestAir because it’s letting them automatically rewrite synopses of their inflight entertainment with new ones that sound weird and aren’t true! (The Chinese vice-premier isn’t even a character in this.)
Aaron Schein tweet media
English
1
0
2
503
Aaron Schein retweetledi
Keyon Vafa
Keyon Vafa@keyonV·
Can an AI model predict perfectly and still have a terrible world model? What would that even mean? Our new ICML paper formalizes these questions One result tells the story: A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws 🧵
English
210
990
6.7K
1.4M
Aaron Schein retweetledi
Jeremy Bernstein
Jeremy Bernstein@jxbz·
When it comes to deep learning optimization, it sometimes feels like people are talking past each other. So, as part of the talk, I decided to survey lots of different ideas and show how they can be put on common footing through a simple algorithmic pattern (4/6)
Jeremy Bernstein tweet media
English
1
2
21
2K
Aaron Schein retweetledi
(((ل()(ل() 'yoav))))👾
we write too much. more than we can read, and many small incremental things. i think there should be some mechanism to restrict paper submissions and acceptances per person per year, to force people to prioritize their best work, and invest more in it.
Jiaxuan You@youjiaxuan

🤯NeurIPS 2025 might break records as the most submitted-to academic conference ever. One of our submission IDs is already ~23,000 — final count could hit 30,000. Absolute madness. #NeurIPS2025 #AI

English
27
29
601
115.7K
Aaron Schein
Aaron Schein@AaronSchein·
Very excited about this work. @JunsolK did an amazing job leading it!
Junsol Kim@JunsolK

#ICLR2025 Thrilled to share that our @iclr_conf paper has been accepted for an oral presentation! We show that LLMs represent political ideologies on a linear axis—and that we can detect and even steer their ideological perspectives. In our paper: - We find a linear "map" of political ideologies within LLM activation space, where U.S. politicians and news outlets are positioned along a left-right spectrum. - By probing this activation space, we can measure the political perspective that LLMs implicitly adopt as they generate text, token by token. - By directly intervening in the activation space, we can steer the model’s output toward more left- or right-leaning perspectives—like turning a dial. Huge thanks to the amazing collaborators and mentors, @profjamesevans and @AaronSchein! 🔗 Full paper: openreview.net/pdf?id=rwqShzb…

English
0
1
12
979
Aaron Schein retweetledi
Junsol Kim
Junsol Kim@JunsolK·
#ICLR2025 Thrilled to share that our @iclr_conf paper has been accepted for an oral presentation! We show that LLMs represent political ideologies on a linear axis—and that we can detect and even steer their ideological perspectives. In our paper: - We find a linear "map" of political ideologies within LLM activation space, where U.S. politicians and news outlets are positioned along a left-right spectrum. - By probing this activation space, we can measure the political perspective that LLMs implicitly adopt as they generate text, token by token. - By directly intervening in the activation space, we can steer the model’s output toward more left- or right-leaning perspectives—like turning a dial. Huge thanks to the amazing collaborators and mentors, @profjamesevans and @AaronSchein! 🔗 Full paper: openreview.net/pdf?id=rwqShzb…
Junsol Kim tweet media
English
1
46
216
26K
Gautam Kamath
Gautam Kamath@thegautamkamath·
At my local cafe, a double shot of espresso is 2.75. An Americano is 3.75 or 4.25, depending on size. Is water that expensive?
English
9
0
28
9.3K
Karl Rohe
Karl Rohe@karlrohe·
Data science like applied statistics is a performance, moving between stances.
Karl Rohe tweet media
English
3
2
10
714
Karl Rohe
Karl Rohe@karlrohe·
Economist @alz_zyd_ has a great critique of data science classes. It’s true and a good thing for us to think and work on. How is data science different from statistics? To many statisticians it isn’t 100% clear. There is something to say beyond “moar applied” ….
alz@alz_zyd_

Because we're better data scientists than the ppl teaching "data science" courses Data science tends to be taught by ppl in stats and applied math departments, who develop tools but spend little time using them, so you learn little about actually doing applied data science

English
3
7
41
15.8K