🌞 Summer of Data Seminar: Diving Deep into Data Research
We're hosting a weekly data seminar series at Datology AI featuring fun and thoughtful researchers pushing the boundaries of pretraining and data curation. Are you data-obsessed yet?
Written by
Share on
Welcome to the Summer of Data Seminar series, hosted by Datology AI. Each week, we invite friends, collaborators, and curious minds doing interesting research in data and pretraining to share what they’re excited about—with a small group of us at the office (or virtually).
Whether it’s how to filter data at scale, reimagine pretraining from scratch, or just what makes a dataset “good”—this seminar is a space to think out loud, ask questions, and hang out with others who are data-obsessed.
🧠About The Seminar
The Summer of Data Seminar is a casual internal series where we bring in folks doing great research around:
- Dataset design, pretraining, or scaling laws
- Synthetic data and data-centric alignment
- Data contamination, memorization, unlearning
- Anything else weird and interesting about data
Each session includes a short talk (30–40 mins) followed by open-ended discussion, questions, and some good old-fashioned geeking out. We record these sessions to share with the broader community on YouTube, while keeping the live discussions cozy with just our team.
The joy of research is in sharing it. And asking the hard questions together.
đź“… Event Schedule
Stay tuned for upcoming talks. Here's who we've hosted so far:
| Presenter | Date | Affiliation | Topic | Resources |
|---|---|---|---|---|
 Charlie Snell | May 5, 2025 | UC Berkeley | Scaling Test-Time Compute & Predicting Emergent Capabilities by Finetuning | |
 Maximilian Böther | May 7, 2025 | ETH Zurich | Mixtera: A Data Plane for Foundation Model Training | |
 Shizhe Diao | May 19, 2025 | NVIDIA Research | CLIMB: CLustering-based Iterative Data Mixture Bootstrapping for Language Model Pre-training | |
 Alexander Gurung | June 2, 2025 | University of Edinburgh | Learning to Reason for Long-Form Story Generation | |
 Jacob Springer | June 9, 2025 | CMU | Echo embeddings & Overtrained Language Models Are Harder to Fine-Tune | |
 Suhas Kotha | June 16, 2025 | Stanford | Standard fine-tuning inefficiently uses rare data | — |
 Xindi Wu | June 23, 2025 | Princeton | COMPACT: COMPositional Atomic-to-complex Visual Capability Tuning | |
 Jaehun Jung | June 30, 2025 | UW & Nvidia | Prismatic Synthesis & G-Vendi Score: How Data Diversification makes R1-32B a Better Teacher than R1-671B | |
 Etash Guha | July 7, 2025 | Stanford & UW | OpenThoughts3 | |
 William Held | July 21, 2025 | Georgia Tech | Optimizing Pretraining Data Mixtures with LLM-Estimated Utility | |
 Vishaal Udandarao | July 25, 2025 | U Tubingen | ACID: Active Data Curation Effectively Distills Large-Scale Multimodal Models | |
 Rosie Zhao | July 28, 2025 | Harvard University | Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining |
🎤 Want to Present?
If you’re working on something fun and would enjoy chatting about it with a bunch of thoughtful nerds, we’d love to have you join us.
To suggest a talk, send us a message (email or DM @datologyai) with:
- Your name + topic
- A rough title
- Any weeks that are particularly good or bad for your schedule
We’ll take it from there.
We’re excited for a summer of curiosity, great conversations, and rabbit holes we didn’t expect to fall into.
Stay data-obsessed. 🤓🚀
Ready for better data?
Let’s make models better through better data, automatically.
Book a Call