The Past as a Compass for Systems Engineers in the Age of AI
summary
A history-driven argument that AI is already outperforming humans in narrow tasks: a LabCorp help‑desk prototype proves ~10% better than five years of human answers; GPQA leaders rise from ~60% to ~84% and HumanEval approaches ~99%. Real incidents include a poisoned embedding (Python pickle) on Hugging Face that escapes to the OS and reaches a shared AWS host, and an Air Canada RAG error costing ~$0.5B market cap. Recommendations: treat observability as evaluations, store models in an artifactory, gate access via Okta and avoid a thousand untracked Jupyter notebooks.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:02Why history matters
- 2:29The historical roots of artificial intelligence
- 8:13Forgotten neural pioneers
- 11:21AI hallucinations and abstraction
- 13:05AI performance in reasoning
- 16:20Shadow IT and AI rise
- 18:47AI technical debt tsunami
- 20:15AI misinformation risks
- 23:04AI security vulnerabilities
- 25:33AI governance and evaluations
- 27:17NORMAL stack for AI
- 30:25Managing AI models and pipelines
- 33:55AI future beyond transformers
- 37:27AI as alien cognition
- 42:45Stop debating AGI
- 43:33Q&A
related talks
21mLearning from History: The Past as a Compass for Systems Engineers in the Age of AI
46mAI vs Humans Rewriting the Rules of Success
01h 06mAI in the Trenches: Engineering Real-World Systems
54mAI vs Humans: Rewriting the Rules of Success
14mHow We Turned AI Hype into Real Value
50mAI IN SRE: HYPE VS REALITY IN 2026
46mCan you trust your AI?
30mIntroduction to AI Safety. Let's be ready before it's too late!