Ryan Holbrook (@ryanpholbrook) profile picture

Data scientist @kaggle / @google. Competitions team.

Brooklyn, NY
Joined December 2010
Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)Photo by Ryan Holbrook (@ryanpholbrook)
Ryan Holbrook retweeted
this is like the first thing you learned when doing kaggle competitions in the good old pre llm days. The person who won did the most experiments. And that person was the person who figured out how to pick the smallest model, data size, experiment which gave them a noisy but still directionally correct and statistically trustable Bayesian update The best research lesson I ever learned was that information gain per unit time is the name of the game
for whatever amount of compute you have, you will get much more research done if you can just make your experiments smaller
27
71
3
1,346
35,995
Ryan Holbrook retweeted
Santa's workshop has a calculator problem. Each toy-making station needs a list of exact numbers, but its Elf Calculator can store only 48 values at a time. Use basic arithmetic (+, −, ×, ÷) to turn small starting values into large targets across 1,000 puzzles, in as few instructions as possible. Keep your program brief, make your logic shine, and help Santa launch right on Christmas time! 🎅 💰 Prize Pool: $50,000 ⏰ Entry Deadline: January 21, 2027
3
10
3
118
11,021
Ryan Holbrook retweeted
ROGII Wellbore Curves Recognition. There are exactly zero ML competitions on Kaggle right now, so I hosted one myself. From Kagglers, for Kagglers.
2
3
31
148,580
Ryan Holbrook retweeted
New #kaggle competition "Santa 2026 - The Elf Calculator" is launched. Medal: True Kernel Only: False Deadline: 2027-01-28 23:59:00+00:00 kaggle.com/competitions/sant…
10
2
90
5,762
Ryan Holbrook retweeted
Want to build the next SWE-bench? Its co-author @OfirPress says to start with people. Find teammates you can bounce ideas off of, then work on really tough problems that seem absolutely impossible. He shares how SWE-bench went from a lukewarm launch to one of AI coding's most influential benchmarks in the first episode of Evaluated, hosted by Gina Lee. Here's a sneak peek:
4
7
1
45
9,547
Ryan Holbrook retweeted
📣 If you have opinions about chart crimes and plotslop, we have an arena for you. Go vote for the best models on Chart Arena, live on Kaggle now! kaggle.com/benchmarks/kaggle…
1
4
19
1,127
Ryan Holbrook retweeted
Interesting write ups are being published now that the competition is closed Many are are inference engineering I want to make all these tricks available to future agents who are experimenting in the same domain How do we contribute this to public knowledge town square?
Competition closed Thank you to everyone who participated - now time to tally it all up V3: 18.5K entrants, 67K submissions V2: 12K entrants, 35K submissions Papers (1 more week): 420 submissions
10
6
50
3,521
Ryan Holbrook retweeted
How do competitors with restricted compute find opportunities to win, build their careers, and earn medals? Last year, Isaiah Pressman finished 2nd in a competition on his home PC, behind a team running 8 H100s. This year, he won on a cluster. In the first episode of Beyond the Leaderboard, three Orbit Wars competitors weigh scale against clever engineering. Here’s a sneak peek:
5
16
5
106
11,741
Ryan Holbrook retweeted
New kaggle logo is pretty dope.
2
3
1
59
2,659
Ryan Holbrook retweeted
As a test of our progress to advance the frontier of AI research, in June we entered the next generation of our autonomous AI research system, AIRA₃, in a live Kaggle competition run by NVIDIA to fine-tune a 30B Nemotron model. The challenge was to teach the model to reason better — all competitors had access to the same information and were graded externally on a private test set. AIRA₃ placed 8th out of ~4,000 teams to win Gold, outperforming human competitors who had access to the same frontier tools. We believe this is a reliable signal that AIRA₃ can improve a targeted capability of an AI model at a level similar to human experts.
109
211
95
1,971
323,064
Ryan Holbrook retweeted
Last month, the NVIDIA Nemotron Model Reasoning Challenge wrapped on @kaggle with 5,000+ participants across 4,000+ teams. The community explored techniques for improving reasoning accuracy, here's what we learned 👇
17
27
4
201
16,418
Ryan Holbrook retweeted
🏆 New @kaggle competition! 🏆 Autonomous Agent Prediction: submit an agent to autonomously solve multiple unseen ML tasks from a common distribution. It's a test of where we're heading -- building & evaluating agents that continually learn, generalizing across novel but related tasks within token budget constraints. kaggle.com/competitions/auto…
3
8
2
71
10,114
Ryan Holbrook retweeted
Deep Past Challenge was my first kaggle competition without coding, and with @raja_biswas we came 3rd 🥉 It was fun to use agentic tools to learn about life and settling disputes 4000 years ago. Thanks @kaggle for hosting 👇10 minas of silver (~5 kg) — enough to buy a house.
4
14
2
101
13,314
NVIDIA is hosting a Kaggle competition. How can you train a nemotron nano model to solve scientific questions? I hope you'll enjoy it! For this competition @kaggle secured NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs from Google Cloud. These GPUs are much more powerful than the usual Kaggle GPUs. Come and try these beasts! kaggle.com/competitions/nvid…
4
72
3
496
31,268
Ryan Holbrook retweeted
1/ Today the biggest score jump in AIMO history happened: we sit at 38/50. Usually score increases are on the order of few points, and it is unusual for the distance to the other competitors to be this big, especially as most of easy problems have now been solved.
3
21
6
179
37,373
Ryan Holbrook retweeted
Shocking results: From 0/50 to 23/50 IMO-level problems in a day! Taking into account that these are very hard problems and your model has on average just 6 minutes of H100-time per problem, this is quite an achievement.
1
4
16
861
Ryan Holbrook retweeted
AIMO3 has launched! Check out our mini-benchmark of reference problems below & help us shrink the gap to commercial LLMs to zero!
2
8
2
29
5,646
Ryan Holbrook retweeted
110 problems created by an international team ❌ Not scraped from the internet ❌ Not in any training dataset ❌ Not seen by any model before This is how you test TRUE reasoning ability. Can your model actually SOLVE problems? Or just pattern spot? AIMO3 will tell you 🔍
1
9
417