DatologyAI published a talk by Nicholas Roberts explaining how smaller models trained on longer data can beat bigger models and change the economics of building AI.
Published
Signal category
Research & Knowledge
Quote
“In this week's Summer of Data seminar, Nicholas Roberts (incoming postdoc at Princeton Language & Intelligence) showed what that insight changes in our process. Once you account for how much a model 'thinks' after training, the most cost-effective build is a smaller model trained much longer on far more data.”
— DatologyAI team
Company
DatologyAI
Better Data, Better Models, Better Business.
- Industry
- Technology, Information and Internet
- Location
- Redwood City, US
- Company size
- 81 employees
DatologyAI builds tools to automatically select the best data on which to train deep learning models. Our tools leverage cutting-edge research—much of which we perform ourselves—to identify redundant, noisy, or otherwise harmful data points. The algorithms that power our tools are modality-agnostic—they’re not limited to text or images—and don’t require labels, making them ideal for realizing the next generation of large deep learning models. Our products allow customers in nearly any vertical to train better models for cheaper.
Founded 2023