Sanket Shah

133 posts

Sanket Shah

Sanket Shah

@sunk8th

AI Scientist @ Duolingo | Prev: CS PhD @ Harvard

Boston, MA Katılım Eylül 2016
1.1K Takip Edilen292 Takipçiler
Sanket Shah retweetledi
Andrew Perrault
Andrew Perrault@PerraultAndrew·
Interested in long-context audio LLMs and hallucinations? We released ~1,140 hrs of synthetic doctor-patient conversations with reference SOAP notes. BeTraC Challenge: build the best open end-to-end SOAP-note system. Two tracks: ≤6B and ≤36B params. betrac.github.io
English
0
4
7
645
Sanket Shah
Sanket Shah@sunk8th·
@GaoZhaolin Can't you suppress the "among responses" variance by, e.g., setting temperature=0? This would give you a cleaner signal for prompt optimization. (I've found that this works quite well in practice.)
English
1
0
0
29
Zhaolin Gao
Zhaolin Gao@GaoZhaolin·
Prompt optimization improves LLMs without updating weights, but can fail when noise dominates the signal. We introduce p1, a prompt filtering method that finds high-signal prompts. Training on just two prompts yields a system prompt that improves AIME 26 from 54.38 to 62.24.
Zhaolin Gao tweet media
English
2
5
35
15K
Sanket Shah retweetledi
Aakriti Kumar
Aakriti Kumar@aakriti1kumar·
✨📢New preprint! Most people feel empathy for others but have a hard time communicating it. We built Lend an Ear, an LLM-powered role-playing platform to help people practice and improve their empathic communication skills.
Aakriti Kumar tweet media
English
2
12
43
7.8K
Sanket Shah retweetledi
Duolingo
Duolingo@duolingo·
Learn Chess on Duolingo Start learning chess for free on Duolingo♟️ Now on Android and iOS Solve bite-sized puzzles and play full games. It's fast, fun, and just a little savage. Ready to make your move? #duolingo #chess
English
80
45
444
59K
Sanket Shah retweetledi
Aakriti Kumar
Aakriti Kumar@aakriti1kumar·
How do we reliably judge if AI companions are performing well on subjective, context-dependent, and deeply human tasks? 🤖 Excited to share the first paper from my postdoc (!!) investigating when LLMs are reliable judges - with empathic communication as a case study 🧐 🧵👇
Aakriti Kumar tweet media
English
2
18
35
9.4K
Sanket Shah retweetledi
AIhub
AIhub@aihuborg·
Interview with Ananya Joshi: Real-time monitoring for healthcare data ift.tt/Pr1wUiz
English
0
1
1
156
Sanket Shah retweetledi
Milind Tambe (Moved @milindtambe-ai.bsky.social)
Milind Tambe (Moved @milindtambe-ai.bsky.social)@MilindTambe_AI·
Huge congrats to PHD student Sanket Shah @sunk8th on his successful PhD defense, "Decision-Focused Learning for the Masses With Applications to Public Health"! 🎉 What a fantastic way to celebrate our Teamcore group's 30th anniversary, with Sanket becoming our group's 40th PhD!
Milind Tambe (Moved @milindtambe-ai.bsky.social) tweet mediaMilind Tambe (Moved @milindtambe-ai.bsky.social) tweet mediaMilind Tambe (Moved @milindtambe-ai.bsky.social) tweet mediaMilind Tambe (Moved @milindtambe-ai.bsky.social) tweet media
English
3
1
26
2.4K
Sanket Shah retweetledi
Sonia Murthy
Sonia Murthy@soniakmurthy·
(1/9) Excited to share my recent work on "Alignment reduces LM's conceptual diversity" with @TomerUllman and @jennhu, to appear at #NAACL2025! 🐟 We want models that match our values...but could this hurt their diversity of thought? Preprint: arxiv.org/abs/2411.04427
Sonia Murthy tweet media
English
3
14
73
7.2K
Sanket Shah retweetledi
ACM Queue
ACM Queue@ACMQueue·
You Don't Know Jack About AI... And ChatGPT probably doesn't either For a long time, it was hard to pin down what exactly AI was. Fast-forward to 2024, and we all now know exactly what AI is. AI = ChatGPT. Or not. queue.acm.org/detail.cfm?id=…
English
0
1
1
726
Sanket Shah retweetledi
Lily Xu
Lily Xu@lilyxu0·
Join us at #ICLR2025 in Singapore! Submit your work at the intersection of machine learning and climate (biodiversity counts!) by Jan 31. We especially encourage submissions that are focused on: 🔢 data-centric methods and challenges 🌏 focused on the Asia / Pacific region
Climate Change AI@ClimateChangeAI

We're excited to announce the next edition of our workshop "Tackling Climate Change with Machine Learning" at #ICLR2025 in Singapore! ▶️ Mentorship program deadline: Dec 27, 2024 ▶️ Paper submission deadline: Jan 31, 2025 Learn more & submit: climatechange.ai/events/iclr2025

English
0
11
63
6.8K
Sanket Shah retweetledi
Panayiotis Danassis
Panayiotis Danassis@PDanassis·
I am thrilled that @_arodriguezca will be giving a keynote talk at the Autonomous Agents for Social Good (#aasg2025) workshop @AAMASconf! Submit your papers by Feb 4th, 2025 and see you in Detroit! More details: panosd.eu/aasg2025/
Panayiotis Danassis@PDanassis

📢Interested in #AIforSocialGood? We invite you to submit any work related to social impact to the Autonomous Agents for Social Good (#aasg2025) workshop @AAMASconf DEADLINE: Feb 4, 2025 See: panosd.eu/aasg2025/ @aparna_taneja @sunk8th

English
0
1
6
269
Sanket Shah retweetledi
Ahmed Alaa
Ahmed Alaa@_ahmedmalaa·
📢 Please retweet: We're recruiting PhD students at UC Berkeley and UCSF! Please apply if you are interests in machine learning for healthcare, statistics, causal inference, or medical vision-language models. For more details, check  out this link: forms.gle/9fbEw48Wqopdfe…
Ahmed Alaa tweet media
English
6
151
386
51.5K
Sanket Shah retweetledi
Santiago Cortés-Gómez
Santiago Cortés-Gómez@sancortes_95·
Excited to share our latest work, where we produce sets with both statistical coverage and high decision utility. Applied to dermatological diagnosis, our method yields sets with coherent diagnostic meaning 🏥. More details in the thread 🧵👇
Santiago Cortés-Gómez tweet media
English
1
13
29
8.2K
Sanket Shah retweetledi
Brandon Amos
Brandon Amos@brandondamos·
📢 My team at Meta is hiring PhD research interns! We study core machine learning, optimization, amortization, flows, and control for modeling and interacting with complex systems (...and we use basic physics... 🙃) Please apply here and message me: metacareers.com/jobs/532549086… 🧵
Brandon Amos tweet mediaBrandon Amos tweet mediaBrandon Amos tweet mediaBrandon Amos tweet media
English
15
90
559
92.5K
Sanket Shah retweetledi
Paula Rodríguez Díaz
Paula Rodríguez Díaz@paularodrid·
🚨 New preprint: How should we measure task similarity when predictions are used for decision-making? Traditional dataset distances based only on features & labels fall short for PtO tasks. Our work with @konglingkai_AI @kaiwang_gua @elmelis @MilindTambe_AI addresses this issue
Paula Rodríguez Díaz tweet media
English
4
14
84
14.1K

Keşfet

@GaoZhaolin @AAMASconf @BondiElizabeth @_arodriguezca @aparna_taneja @TomerUllman @jennhu @konglingkai_AI