Amazing things that happen with Human-AI synergy
summary
Researchers show graph neural networks embedded in Google Maps yield travel-time predictions that cut error by over 40% in some cities (Sydney). A GNN flags an unexpected molecule halicin as a 99%‑confident antibiotic and lab tests confirm potent activity against resistant bacteria. Applied to pure math, GNN saliency guides conjectures: a natural slope approximates knot signature (refined using injectivity radius to prove a bound) and a representation‑theory model jumps from 65% → 99% accuracy after adding a “diamond” subgraph feature, enabling new theorems.
AI-written summary · about
chapters (10)
- 0:00Human-AI synergy intro
- 1:31What AI means today
- 8:36Interrogating neural networks
- 8:57Graph neural networks in systems
- 13:27AI-discovered antibiotics
- 17:58AI for new math theorems
- 22:09Geometry meets algebra in knots
- 27:24AI solves 50-year math challenge
- 35:08Future of explainable AI
- 35:50Q&A
related talks
39mUsing Graph Neural Networks to Embrace the Dependency Within Your Data
16mPreview: Graph NNs
01h 01mToward Universal Graph Representations: Foundations and Frontiers of Graph Foundation Models
01h 20mComparing GNN Approaches for Molecule Identification
29mGraph Neural Networks + LLM on
48mKeynote – Sameer Singh
23mUsing Graph Nets (GNs) to predict molecular properties
23mGNN-RAG: Bridging Graph Reasoning and Language Understanding for Knowledge Retrieval