Log inSign up
DatologyAI
226 posts
user avatar
DatologyAI
@datologyai
DatologyAI builds tools to automatically select and optimize the best data on which to train AI models, leading to better, smaller models which train faster.
Redwood City, CA
datologyai.com
Joined September 2023
10
Following
3,075
Followers

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

TermsยทPrivacyยทCookiesยทAccessibilityยทAds Infoยทยฉ 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up
  • user avatar
    DatologyAI
    @datologyai
    23m
    The hottest release at ICML isnโ€™t just another datasetโ€”itโ€™s our newest fidget spinner. Weโ€™ve upgraded this yearโ€™s design, and theyโ€™re waiting for you at the DatologyAI booth. If youโ€™ve been collecting them throughout the year, you know the drill: gotta catch โ€™em all. Stop by,
    264
  • DatologyAI reposted
    user avatar
    Ari Morcos โœˆ๏ธ ICML
    DatologyAI
    @arimorcos
    21h
    Just landed in Seoul for #ICML2026! Please reach out if you're interested in chatting about all things data, from filtering to scaling to synthetic, or if you'd like to learn more about working with us @datologyai to understand the fundamental principles of data curation!
    2.5K
  • user avatar
    DatologyAI
    @datologyai
    Jun 30
    Bogdan Gaza @hurrycane on what it takes to run a 20T-token data pipeline, live at @aiDotEngineer. Booth U-G3 to go deeper.
    2.1K
  • user avatar
    DatologyAI
    @datologyai
    Jun 30
    "If you choose your data correctly, you can bend the scaling laws. Data curation is the lever." @arimorcos live at @aiDotEngineer right now. Missed it? Come talk data quality at Booth U-G3.
    1.3K
  • user avatar
    DatologyAI
    @datologyai
    Jun 30
    Today at @aiDotEngineer. Two talks from the DatologyAI team: "Data Quality is the Compute Multiplier," Ari Morcos, 10:45am, Room 2016 "Running a 20T-Token Data Pipeline," Bogdan Gaza, 3:20pm, Expo Stage 3 SE. Booth U-G3 all day. @arimorcos @hurrycane
    477
  • user avatar
    DatologyAI
    @datologyai
    Jun 29
    This Tuesday at The AI Engineer Worldโ€™s Fair in SF @aiDotEngineer, our co-founder & CEO @arimorcos takes the Data Quality stage for the keynote. "Data Quality is the Compute Multiplier": why better quality training data drives a better model with the same compute. Tue 6/30, 10:45
    2.6K
  • user avatar
    DatologyAI
    @datologyai
    Jun 26
    The first talk from the Summer of Data 2026 is live. Thanks to Konwoo Kim and Suhas Kotha (Stanford) for stopping by to talk about synthetic megadocs and squeezing more out of training compute. Full talk in the reply. @konwookim @kothasuhas
    00:00
    3.5K
    user avatar
    DatologyAI
    @datologyai
    Jun 26
    Full seminar talk here:
    394
  • user avatar
    DatologyAI
    @datologyai
    Jun 26
    Summer of Data is back for 2026. We invite researchers to our office to talk to our team about their work on data and model training, and then we post the talks so you can watch. Speakers from Stanford, MIT, NVIDIA, CMU, NYU, UMD, Tรผbingen, Wisconsin, and more. We are posting a
    1.4K
    user avatar
    DatologyAI
    @datologyai
    Jun 26
    Check the blog for updates each week:
    datologyai.com
    ๐ŸŒž Summer of Data Seminar 2026: Deep in Data Research
    We're hosting a weekly data seminar series at Datology AI featuring researchers pushing the boundaries of training and data curation. Are you data-obsessed yet?
    447
    user avatar
    DatologyAI
    @datologyai
    Jun 26
    Or follow our YouTube channel
    youtube.com
    DatologyAI
    At DatologyAI, we are focused on building the worldโ€™s best platform for automated data curation at petabyte scale, helping customers to train models faster, to better performance, and enabling...
    279
  • user avatar
    DatologyAI
    @datologyai
    Jun 25
    New @datologyai research: 35ร— cheaper per correct answer through pretraining data curation alone. No parameter reduction or decoding tricks required. ๐Ÿงต A million correct answers cost ~$1.34 from our curated 4B, and ~$47 from verbose Qwen3.5-4B. Same answers.
    01:20
    user avatar
    Matthew Leavitt
    DatologyAI
    @leavittron
    Jun 25
    What if you could induce models to be more concise via pretraining data curation?
    1.4K
    user avatar
    DatologyAI
    @datologyai
    Jun 25
    Replying to @datologyai
    Thoughful pretraining curation allows the model to learn to be concise, so brevity is built into the weights, not bolted on at decode time.
    183
    user avatar
    DatologyAI
    @datologyai
    Jun 25
    ๐Ÿ“„ Paper: arxiv.org/abs/2606.25432 ๐Ÿ“ Blog: datologyai.com/blog/brevity-iโ€ฆ Join us: datologyai.com/careers Become a customer: datologyai.com/contact
    111
  • user avatar
    DatologyAI
    @datologyai
    Jun 22
    The "you can only catch up by distilling from a frontier model" narrative is wrong. We curated the data for @Arceeai's Trinity Large entirely from public sources, zero closed-model APIs, and it's competitive with the open frontier. Better data does the work.
    00:00
    1.2K
    user avatar
    DatologyAI
    @datologyai
    Jun 22
    Full episode:
    268
  • user avatar
    DatologyAI
    @datologyai
    Jun 19
    Compute scarcity is about to force the reckoning the frontier labs have avoided: efficiency. You don't need trillion-parameter models for frontier-class capability. With better data, far smaller models match the best of a year or two ago, at a fraction of the cost to serve.
    00:00
    1.1K
    user avatar
    DatologyAI
    @datologyai
    Jun 19
    Full episode:
    645
  • user avatar
    DatologyAI
    @datologyai
    Jun 18
    Replying to @datologyai
    5/ Alexander Gurung from the University of Edinburgh presented his work on learning to reason for long-form generation. What does a reward signal look like when the goal is a good story? ๐Ÿ“บ youtu.be/tB8dx9QGVcM
    418
    user avatar
    DatologyAI
    @datologyai
    Jun 18
    Replying to @datologyai
    14/ Sukjun Hwang (@sukjun_hwang) from CMU presented his work on H-Nets: Dynamic chunking for end-to-end hierarchical sequence modeling ๐Ÿ“„
    arXiv logo
    arxiv.org
    Dynamic Chunking for End-to-End Hierarchical Sequence Modeling
    Major progress on language models (LMs) in recent years has largely resulted from moving away from specialized models designed for specific tasks, to general models based on powerful architectures...
    225
    user avatar
    DatologyAI
    @datologyai
    Jun 18
    15/ What a lineup, and that was only year one. Summer of Data is back for 2026 and we're just getting started. Keep an eye out for our lineup announcement and new talks every week. Want to present? DM us ๐Ÿ‘€ Stay data-obsessed ๐Ÿค“
    198