Full house at FAIR's NYC 'Synthetic Data for Reliable AI' event
thegautamkamath · x · 2026-11-17
The "Synthetic Data for Reliable AI" event in NYC drew a full house, kicked off by FAIR's Mark Ibrahim and Karen Ullrich as the opening of an AI talk series.
Speakers included Gautam Kamath, Shubham Toshniw, and several PhD students. Slides covered synthetic data for privacy: data that behaves like the original without revealing individual datapoints, where a sensible first step is fitting a generative model on the original dataset and sampling from it — though synthetic data votes wrongly on unseen dimensions and can yield 5% completely wrong samples. A running joke: even world models research has an East Coast–West Coast rivalry.
More from Companies & People
- SemiAnalysis: over 1.5 GW of non-compliant chips in China; NVIDIA's exit could sink open-source AI — teortaxesTex · 2026-11-17
- The 'Darkest Moment for Chinese AI': Compute, Distillation and Robotics All Under Pressure — PAstynome · 2026-11-17
- A ~50K-seat enterprise wants to fine-tune its own models on Plane data — JosephJacks_ · 2026-11-17
- NUS Computing names nine inaugural Ascent Fellows for early-career researchers — rishabh16_ · 2026-11-17
- Sean Linehan joins Hadrian as VP Product to build AI-run factories — PaulYacoubian · 2026-11-17
- Anthropic to give all employees an executive-style stock trading plan after IPO — terryyuezhuo · 2026-11-17