AI/ML Won't Save you
summary
Argues that ML systems repeatedly fail at real ops problems and gives concrete replacements: a 2011 deployment uses a probe mesh over Force10/MPLS paths plus a linear programming engine to pinpoint a single shared component causing correlated packet loss of fractions of a percent, not an NN black box. Syslog and firehose anomaly hunting produce vast false positives; operations instead use simple tools like logcheck, fuzzy matching of the top 100 events, and intent/discovery data to reduce noise.
AI-written summary · about
chapters (4)
related talks
25mAutomating Operations with ML
32mAI/Machine Learning: What is Actually Working in Cybersecurity?
38mHow AI Improves Operational Effectiveness, Safety and Well-Being
18mAI Observability: How To Fix Issues With Your ML Model
30mThe Robots Are Coming (But Not Yet), or, Don't Make Smart People Do Dumb Things
10mML-based Log Analysis for Faster Debugging
49mAn AI with an Agenda
27mAI-Powered Reliability: Stop Fighting Fires, Prevent Them