AIRe - AI Reliability Engineering
summary
Argues that reliability for AI means applying SRE practices to models and pipelines rather than treating AI as an extra layer. Demonstrates concrete work: an AutoGPT demo and a custom Longchain agent running GPT‑4 Turbo Preview instrumented with OpenTelemetry and the Open Inference tracer feeding a Phoenix UI, producing traces and metrics for debugging. Presents adoption and production statistics (LM growth; “one in three” experimental models reaching production) and operational challenges—7500‑node Kubernetes at GPT‑3.5 scale, data‑poisoning and GDPR risk, and cost/latency tradeoffs.
AI-written summary · about
chapters (7)
- 0:00Intro. Agenda
- 3:57Inspiration
- 6:58Numbers
- 8:00AI Current State
- 11:55AIRe Challenges
- 14:53AI Observability
- 26:44Q&A
related talks
44mAgentic Observability — The Path to AI Co-SRE
26mCan SREs become AI/ML engineers?
16mHow AI is redefining SRE and Customer Experience
01h 00mAI A-Z
26mbuilding reliable AI systems in production
44mSRE and ML: Why It Matters
47mEmbedding AI: Don’t Burn Your Fingers in the Heat of the AI revolution
46mCan you trust your AI?