Post #1866

Hugging Face (Twitter) RT @ClementDelangue: I believe open datasets are becoming the most important contributions in AI. A few reasons: - We still lack truly great open datasets across domains, modalities, languages, and techniques - and the gap is especially painful in reinforcement learning - Datasets are costly, require a lot of hard work, and deeply unsexy, which makes them hard to create for smaller teams. Without strong open datasets, the open ecosystem simply can’t compete on equal footing with closed labs - Open datasets gives you a whole new layer of transparency and replicability that allows you to study biases, efficiency, interpretability and much more. - Open datasets compound in value. You can reuse them with every new architecture, optimizer, or training breakthrough, turning a single dataset into hundreds of state-of-the-art models over time. Models depreciate whereas open datasets appreciate. That’s why I... Перейти на оригинальный пост