Josh Wills, Head of Data Engineering, Slack
summary
Model building time collapses from months to days using XGBoost, embeddings, and emerging neural architecture search tools, enabling seismic oil imaging that used to take months to run in a week and rapid experiments on spam, ad click, and revenge‑porn detection. The hard problem is product and policy: data needed to solve many harms sits scattered across distrusted platforms, so the proposed fix is small, shared datasets and fair collective‑action mechanisms (start with security and build toward health) to enable trustworthy, auditable models.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:35What I do now
- 1:53My last decade
- 2:12Google
- 4:04XG Boost
- 4:57Embedding
- 5:51State of the art
- 7:01Archived sanity
- 7:46going further
- 8:53Google ML
- 10:07Technology
- 10:54Data Product Problem
- 13:39Chicken and the Egg
- 15:05My Concern
- 16:08I Wear the Black Hat
- 18:01Trust Issues
- 20:20Collective Action Problems
- 21:23Start Small
- 22:50Information Security
related talks
29mSeveral People Are Tuning: Data and Machine Learning at
14mBuilding Machine Learning Models with Strict Privacy Boundaries
42mXavier Amatriain, VP of Engineering, Quora
33mPanel Discussion
35mBuilding Data-Intensive Teams
39mDataEngBytes 2023 - BRI-T1-02 - Mikiko Bazeley
27mNavigating Data's Next Great Shift
46mWhy data-driven methods will shape the future (...)