AI Won’t Help You Here
summary
Generative models are often the wrong tool: a Slack-trained assistant parrots workplace habits ("I'll do it tomorrow"), Copilot wastes thousands on wrong home-automation answers, and LLMs hallucinate facts (claiming a watermelon is red outside and green inside). Deterministic systems work: the aviation traffic collision avoidance system uses decision trees and vectors and prevents crashes. The practical fix is narrow data + factual models — knowledge graphs, decision logic, documentation-trained parsers — exemplified by a released no excuse code fix that produces deterministic, verifiable patches.
AI-written summary · about
chapters (9)
- 0:00Introduction to AI talk
- 2:10Generative AI misuse examples
- 9:44Security basics vs AI hype
- 10:50Positive AI use cases
- 14:51AI for unstructured data
- 17:18Flaws in generative AI
- 20:08AI for code vulnerabilities
- 23:19Deterministic models over generative
- 28:41Q&A session
related talks
55mAI Showdown: The Good, The Glitchy, and The Risky
44mToxic Trust: The Psychology of Automation Bias in Generative AI
32mBeyond LLMs: Building an AI-Powered Decision System at Scale
52mAI Is Just Software, What Could Possibly Go Wrong?
50mFrom Prompt Hacking to Architectural Determinism: Engineering Reliable GenAI Systems
16mBuilding Applications with AI Agents
17mGenerative AI: Keeping it real
45mThree Practical Ways to Use AI in Your Company