@maharshigor

Ph.D. student @umdcs @ClipUmd NLP, Retrievers, Human-AI, Efficient Methods Past: @ContextualAI @Cohere @GoogleAI @theteamatx 🚀 he/him 🏳️‍🌈

 College Park, MD
mgor.info
Joined January 2011
  • Tweets344
  • Following615
  • Followers430
25Media
Can you spot when AI bluffs?🤖 Can you outguess AI—or work with one to dominate trivia?🏁 🏆 We are hosting the first Human–AI coop trivia (Quizzing) competition. 🎲Play, 🛠️build, or ✍🏼write questions... ..and win prizes 🎁. 🥳 It’s fun, free, and happening this June 🧠🤖👇
1
15
3
19
3,557
Maharshi Gor @ ACL2026 🏝️ retweeted
Nishant is at EMNLP@NishantBalepurOct 19
🚨 New Paper! 🚨 Multiple-choice benchmarks use accuracy (boring), but education has way cooler scoring schemes! 🤩 👎 Lose points if wrong 🔁 Try until you're right 🃏 Wager points At #EMNLP2026, we show these schemes shift LLM ranks, align with LLM Arena, + unmask LLM quirks
5
21
1
95
3,413
Maharshi Gor @ ACL2026 🏝️ retweeted
Nishant is at EMNLP@NishantBalepurSep 28
🚨 New Paper! 🚨 Your agent is writing code faster than you can understand it 😬 In our controlled online study, coding agents helped users finish tasks quickly, but reduced how well they understood the code they just “wrote” We also find three ways to avoid this tradeoff 👇🧵
5
40
2
128
9,623
Maharshi Gor @ ACL2026 🏝️ retweeted
Voice Arena
@voicearena_aiJul 3
Introducing Voice Arena STT Leaderboard. LibriSpeech is older than the transformer. FLEURS is people reading translated Wikipedia sentences aloud. This is still how the industry ranks speech models. In 2026. Today we're releasing the Voice Arena STT Leaderboard - a human-verified benchmark for STT accuracy on real, spontaneous conversation, across 7 languages: 🇺🇸 US English, 🇮🇳 Hindi, 🇧🇩 Bangla, 🇧🇷 Brazilian Portuguese, 🇻🇳 Vietnamese, 🇷🇴 Romanian & 🇸🇦 Arabic 🧵 US English Leaderboard ⬇️
4
100
2
76
4,304
📍 Presenting this on July 5 at 11:00 AM in Poster Session A at #acl2026nlp. 📜: arxiv.org/pdf/2605.28255 Come chat about when humans should let AI take the wheel, when they should push back, and what live human-AI collaboration reveals that static benchmarks miss.
Maharshi Gor @ ACL2026 🏝️
@maharshigorJul 5
🧵 Trusting AI is not one decision. In our ACL 2026 paper, expert trivia players had to decide when to let AI answer, when to ignore it, and when to change their minds because of it. The main failure mode was not blind trust.
2
9
726
Maharshi Gor @ ACL2026 🏝️ retweeted
Jordan Boyd-Graber@boydgraberJul 5
We made a trivia game that has humans and computers cooperate: humans could let AIs buzz in for them to answer trivia questions (complete delegation) and then they worked with computers to answer other trivia questions collaboratively. Come to #acl2026nlp Poster Session A!
3
3
11
805
🧵 Trusting AI is not one decision. In our ACL 2026 paper, expert trivia players had to decide when to let AI answer, when to ignore it, and when to change their minds because of it. The main failure mode was not blind trust.
2
3
1
5
1,141
more replies
🛠️ We distill 5 guidelines for human-AI systems
1
1
27
🙏 Huge thanks to everyone without whom this wouldn't be possible: my amazing co-authors: @YooYeonSung1, @houyu0930, @enfleisig, Irene Ying, @zhoutianyi, and @boydgraber; the UMD CLIP group; all participants and question writers/editors.
1
39
Maharshi Gor @ ACL2026 🏝️ retweeted
John Kirchenbauer@jwkirchenbauerMay 25
Earlier this month I successfully defended my PhD and graduated from UMD! 🐢 Thanks to everyone who played large or small role in making these last 5 years an amazing experience. Next stop, Toronto! Defense recording: youtu.be/j8YIv25ge9Q
5
7
62
3,414
Maharshi Gor @ ACL2026 🏝️ retweeted
Jordan Boyd-Graber@boydgraberMay 20
Our work on human-computer QA is the cover of UMD's Enterprise Magazine, featuring the work of @YooYeonSung1 , @enfleisig , @maharshigor, @h__j___han, @EntilZhaPR , @ihsgnef, @MohitIyyer (and many others). Links below. We need people for next competition!
1
5
15
1,013
Maharshi Gor @ ACL2026 🏝️ retweeted
Yapei Chang
@YapeiChangApr 30
How2Everything will appear in ICML 2026! See you in Korea 🫡 We mine the web's procedural knowledge to better evaluate & train LLMs to generate valid step-by-step instructions, read more at: 🔗 arxiv.org/pdf/2602.08808
Ai2
@allen_aiFeb 10
LLMs often generate step-by-step instructions, from real-world tasks (how do I file taxes?) to plans for AI agents. Improving this is hard: outputs can sound fluent for steps that don't work, and current datasets cover few domains. How2Everything evals/trains for this at scale. 🧵
7
1
71
10,662
Maharshi Gor @ ACL2026 🏝️ retweeted
Jordan Boyd-Graber@boydgraberApr 21
Can you spot when AI bluffs? Can you outguess AI or work with one to dominate trivia? Can you make an AI that can figure out tricky visual questions?
Logo of a human and a computer looking at a Korean painting together, advertising the ICML 2026 workshop.

ALT Logo of a human and a computer looking at a Korean painting together, advertising the ICML 2026 workshop.

1
14
2
19
9,977
Maharshi Gor @ ACL2026 🏝️ retweeted
Nishant is at EMNLP@NishantBalepurApr 7
Super grateful to have four papers at #ACL2026 main! Thanks to all my co-authors from UMD, NYU, Ai2 + more 🎉🥹 Expect lots of "🚨 NEW PAPER" posts coming from me soon 😅
3
3
1
90
4,210
Load more