Kiran Garimella
@gvrkiran
Followers
4K
Following
10K
Media
214
Statuses
2K
Assistant Professor at Rutgers School of Information and Communication.
Joined October 2008
Today, I'm releasing videos in the series on how I use AI in my life. The series is a collection of videos on how i use AI tools like ChatGPT, Claude and Gemini for research, ideation, writing, image/video/audio generation, data collection, gaming, etc. https://t.co/RP5y7qdxeV
youtube.com
Various things I do with AI. for research, teaching, fun, etc.
4
14
100
This paper shows we can fine-tune LLMs on convenience samples to recover accurate population-level distributions So instead of complete silicon samples, you can get convenience samples and "learn" population level traits (unlike traditional reweighting) https://t.co/U9l1uzaBZM
0
0
2
UC Berkeley EECS is hiring! We're seeking exceptional faculty candidates at all ranks for our "Engineering + AI" search and up to 7 tenure-track Asst. Professors in EECS. EECS Focused Searches Include: Quantum Computing ⚛️ AI, Inequality, & Society ⚖️ https://t.co/S8KFqbx1yF
eecs.berkeley.edu
Explore open faculty positions at UC Berkeley's Electrical Engineering & Computer Sciences (EECS) Department. Join a world-class research and teaching community. Apply today.
1
26
100
This is SO impressive and important
🚨 out at @apsrjournal 🚨 ➡️ We ran a large media literacy experiment to fight misinformation ➡️ 13,500 students, 583 villages in Bihar, India ➡️ Created custom misinfo curriculum of 4 months ➡️ Partnered w the government to roll it out as an official course in classrooms
1
0
11
This paper answers this question: "we find that left-leaning documents consistently outnumber right-leaning ones by a factor of 3-12 across training datasets, with pre-training data having abt 4x more politically engaged content than post-training data." https://t.co/APbIxjtiq6
This study shows how simplistic methods overstate biases in chat bots. Their measure reveals a subtler leftward tilt in instruction-tuned language models. Prompting methods matter This bias will always exist probably because the internet is left leaning. https://t.co/ldvTFpixJh
0
1
10
I'll be in Boston 15-17 (Wednesday-Friday) attending an event at Harvard. Would love to meet and chat if you're around.
0
0
2
An audit of LLMs in college admissions found they favor low-income applicants, even with equal academic scores. Raises a Q: We focus on removing bias from AI, but what do we do when AI develops an "accepted" bias that mirrors deliberate social policy? https://t.co/6rwHaN3lVM
0
0
2
A dataset of 1000 food recipies from rural parts of India https://t.co/E95zrXpA1J The dataset may not be interesting to many but to me its interesting that we now are creating these datasets for AI. We'll see more such documenting humanity for a non-human audience.
0
6
31
The AI, Interaction, and Learning team at Microsoft Research is looking for interns! If you're working on your PhD in computer science, statistics, economics, computational social science or related fields apply: Research Intern MSR AI Interaction and Learning | Microsoft Careers
1
6
50
Tools for detecting LLM text: https://t.co/jkEXHBPKfg Claims to be 70% better than prev work. I personally think such tools are brittle and dont work well because LLM text is now as good/better than humans. But still, we need such tools for studying the prevalence of LLM use.
1
0
1
the deadline for this position is Oct 15. lmk if you have any questions!
Our department at Rutgers is hiring for a position in Information Policy this year. details here: https://t.co/Ls1IxzMeOI
0
1
2
We’re recruiting at Johns Hopkins School of Government and Policy! We’re looking for (among others) people with interests in AI, Science and Innovation, coming from any relevant disciplinary background. One noteworthy point: we’re keen for people with CS PhDs to apply.
9
104
258
Whenever we see such “enragement bait” posts, our prior should be that these are AI generated.
I just picked up my daughter from her Canadian school wearing her beautiful American flag sweater. I didn’t know when taking this picture from my car, but I have just learned that a teacher asked what she was doing wearing this today, and literally accused her of supporting a
1
0
1
I wrote some thoughts about the "slow down" in AI progress and why it might have to do mostly with costs than actual progress. https://t.co/nX8qHrnvSL
0
0
1
This paper looks at the use of LLMs in news articles. They find that at most 1% of the articles could have used GenAI (with a variance on where they used it). Tools to detect genAI content are not great so this might be an undercount but its a good sign https://t.co/vkffxj3ual
0
0
2
This paper shows that AI propaganda has moved from a future possibility to a current reality. They use small language models on commodity hardware to show current capabilities. Their solution: move from restricting model access to detection & disruption https://t.co/T3hcztYJJ2
0
2
5
AI is already outperforming humans in tasks like coding, yet we still insist on micromanaging it. I wrote about the challenge we might face pretty soon of learning to let go. https://t.co/Gprcfc5qfH
0
1
4
This paper from Google applies vision language models to all streeview data to make it interactive. Particularly useful for visually challenged users (among many other uses). Cool idea! Hope it gets released as a product. https://t.co/FjKxODWyZy
0
0
2
Educators have been using AI for developing exams (we just dont talk about it). e.g. see this great Claude report https://t.co/vyfQKrNCMk How good are those AI generated exams? This paper shows they work great, comparable to human generated exams. https://t.co/mJnT40YN7K
0
0
3