Panel Discussion
summary
Panelists argue that tooling is transient so prioritize the data and model life cycle over specific libraries; treat LLM deployments as ML problems with familiar risks not magic. Practical steps include running a spreadsheet tradeoff of latency versus accuracy, instrumenting logging monitoring alerting, and selling changes as revenue or risk reduction rather than cost cuts. Advice: focus on grounded use cases, expect messy corporate tabular data, and require minimal guardrails before shipping to production.
AI-written summary · about
chapters (13)
- 0:00Welcome and panel
- 1:05Mickey introduces himself and shares Australian observations
- 2:30Joe introduces himself and shares Australian coffee insights
- 4:16ML Ops discussion
- 6:04Lifecycle thinking
- 8:08Breaking silos
- 11:20AI hype question
- 12:26Evaluating AI trends
- 17:44Corporate data challenges
- 19:08Cost vs revenue
- 24:02Delivery vs quality
- 26:13Risk assessment
- 30:48Closing remarks
