Research @StanfordAILab Retrieve knowledge for better decision making
Palo Alto, CA
- Tweets 77
- Following 127
- Followers 718
- Likes 460
Pinned Tweet
AI can increasingly make progress on open problems like Navier-Stokes. Yet defining a new problem still relies on human research taste.
Introducing ScholarCatalyst, a benchmark built from AI researchers’ firsthand accounts of what inspired their work.
5
42
208
19,028
Sohyeon Kim @ COLM 2026 retweeted
Really happy to contribute my recent NeurIPS 2025 paper to ScholarCatalyst!
I think this is a really important step toward AI scientists that can learn from the broader research space and identify prior ideas that genuinely enable new discoveries.
AI can increasingly make progress on open problems like Navier-Stokes. Yet defining a new problem still relies on human research taste.
Introducing ScholarCatalyst, a benchmark built from AI researchers’ firsthand accounts of what inspired their work.
1
2
7
695
Sohyeon Kim @ COLM 2026 retweeted
Measuring research taste in what inspires new research ideas … check out our new benchmark…
AI can increasingly make progress on open problems like Navier-Stokes. Yet defining a new problem still relies on human research taste.
Introducing ScholarCatalyst, a benchmark built from AI researchers’ firsthand accounts of what inspired their work.
1
2
29
2,266
Sohyeon Kim @ COLM 2026 retweeted
To tackle new scientific problems, we often get inspirations from prior work.
Can Deep Research agents and search help us to identify such “catalyst” papers buried in literature?
Our new benchmark, built with hundreds of scientists, shows substantial room for improvements.
AI can increasingly make progress on open problems like Navier-Stokes. Yet defining a new problem still relies on human research taste.
Introducing ScholarCatalyst, a benchmark built from AI researchers’ firsthand accounts of what inspired their work.
4
9
63
5,349
AI can increasingly make progress on open problems like Navier-Stokes. Yet defining a new problem still relies on human research taste.
Introducing ScholarCatalyst, a benchmark built from AI researchers’ firsthand accounts of what inspired their work.
5
42
208
19,028
One final note to end on: as @arxiv reported yesterday, it now has 3M+ submissions, received 40K+ last month alone, and cs.AI submissions have grown 6X+ in just two years.
Finding the few papers that can actually move your research forward is *only* getting harder, and remains part of the research taste that AI has yet to match.
x.lingyaoai.com/arxiv/status/210573535…
arXiv has updated our policy on rate limiting for all submitters.
This update was made to fairly distribute moderator time & support the arXiv community of staff, volunteers, readers & authors.
Please read our announcement to learn more: blog.arxiv.org/2026/10/01/up…
ALT arXiv logo in white over a whiteboard with math symbols.
1
5
247
Also check out @yoonholeee's thread for a deeper discussion of the ideas behind this work and what we mean by research taste!
Great researchers have an uncanny ability to make connections that seem inevitable in hindsight, in places nobody else would have thought to look. Can we measure this ability?
Introducing ScholarCatalyst: a far-from-saturated benchmark for finding what we call "catalyst papers"📚, labeled by 184 lead authors on 207 of their own recent projects.
Paper: arxiv.org/abs/2610.02202
To make sustained progress on open-ended problems, I think agents need the sort of "research taste" that great researchers have. They need to make deep connections between earlier discoveries and problems those discoveries weren't intended to solve. ScholarCatalyst is a first step towards this goal.
More details in the thread below🧵
4
261
Sohyeon Kim @ COLM 2026 retweeted
I expect a substantial number of scientific breakthroughs to come from taking insights from one field and applying them to another. Very few people get PhDs in multiple areas -- there just isn't time for it in most human lives. A model can get a PhD in every field though (thanks, parallelization!).
When my wife and I dated near the end of our undergrads, we discovered that we had learned very similar concepts with totally different terminology in engineering vs economics programs (shadow price? you mean Lagrange multiplier?).
I look forward to all the accelerated insights we'll have as a species once relevant work is surfaced to researchers, regardless of what field it's coming from. There are Nobel prize sized contributions already in reach for the person who discovers that the intractable problem in field A has already been solved in field B. Scholar Catalyst is a great first step in that direction. Proud to support it as a @LaudeInstitute Slingshot!
Great researchers have an uncanny ability to make connections that seem inevitable in hindsight, in places nobody else would have thought to look. Can we measure this ability?
Introducing ScholarCatalyst: a far-from-saturated benchmark for finding what we call "catalyst papers"📚, labeled by 184 lead authors on 207 of their own recent projects.
Paper: arxiv.org/abs/2610.02202
To make sustained progress on open-ended problems, I think agents need the sort of "research taste" that great researchers have. They need to make deep connections between earlier discoveries and problems those discoveries weren't intended to solve. ScholarCatalyst is a first step towards this goal.
More details in the thread below🧵
1
9
58
6,191
Sohyeon Kim @ COLM 2026 retweeted
Research taste will matter more and more as AI starts doing research on its own. Part of taste is knowing which old papers to build on. ScholarCatalyst is a first step toward measuring that automatically :))
Great researchers have an uncanny ability to make connections that seem inevitable in hindsight, in places nobody else would have thought to look. Can we measure this ability?
Introducing ScholarCatalyst: a far-from-saturated benchmark for finding what we call "catalyst papers"📚, labeled by 184 lead authors on 207 of their own recent projects.
Paper: arxiv.org/abs/2610.02202
To make sustained progress on open-ended problems, I think agents need the sort of "research taste" that great researchers have. They need to make deep connections between earlier discoveries and problems those discoveries weren't intended to solve. ScholarCatalyst is a first step towards this goal.
More details in the thread below🧵
2
7
39
2,125
Sohyeon Kim @ COLM 2026 retweeted
Great researchers have an uncanny ability to make connections that seem inevitable in hindsight, in places nobody else would have thought to look. Can we measure this ability?
Introducing ScholarCatalyst: a far-from-saturated benchmark for finding what we call "catalyst papers"📚, labeled by 184 lead authors on 207 of their own recent projects.
Paper: arxiv.org/abs/2610.02202
To make sustained progress on open-ended problems, I think agents need the sort of "research taste" that great researchers have. They need to make deep connections between earlier discoveries and problems those discoveries weren't intended to solve. ScholarCatalyst is a first step towards this goal.
More details in the thread below🧵
16
47
297
20,576
Sohyeon Kim @ COLM 2026 retweeted
‼️The Bitter Lesson for context management: Giving LMs unrestricted control over their context beats human-designed SOTA!
Introducing 🩵Context Language Models (CLMs)🩵
- Natively manage their own context
- Treat context as a file
- Learn policies in CLM weights, no harness
89
266
2,065
428,386