EgoTools: 100-hour egocentric tool-use dataset lifts an 8B model's accuracy by 10.9 points
liuziwei7 · x · 2026-10-04
NTU S-Lab, with ASTAR, KAIST and others, released EgoTools, probing whether AI models can reason about tools the way humans do—not just what a tool is, but what it can do and when to use it.
- Dataset: 100.37 hours of real-world first-person video across 7 everyday domains, from kitchens to research labs
- Benchmark: 1,000 diagnostic questions across four tracks, testing tool grounding, usage tracking, and choice inference
- Model: an 8B reference model improves overall accuracy from 50.0% to 60.9% (+10.9 pp) with the same backbone and 64 frames
- Paper, code, data (Hugging Face), and the EgoTools-8B model are all publicly available
Related event: EgoTools: 100-Hour Egocentric Dataset for Tool-Use Reasoning(4 posts)→
More from Research
- Paper at NeurIPS: individual parameters in weight-sparse transformers appear interpretable — CatAstro_Piyush · 2026-10-04
- How vector databases work: embeddings, cosine similarity, HNSW, IVF and PQ explained — blaizedsouza · 2026-10-04
- CoFlow adds contrastive trajectory repulsion to Flow Matching, improving few-step generation on ImageNet — burny_tech · 2026-10-04
- ARROW unifies 3D reconstruction and point tracking from arbitrary image sets, sets new SOTA — CSProfKGD · 2026-10-04
- Latent communication papers: KV-cache integrity vs malicious agents and state-delta hybrid channels — burny_tech · 2026-10-04
- DNA design trick: fit with gradient descent first, then snap to manufacturer catalogue — anshulkundaje · 2026-10-04