Improving Credibility of ML Solutions
summary
Claims concrete fixes: Motherbrain at EQT Partners fuses two employee data feeds and fuse two data sources with sliding-window ratio to beat deep learning time-series models and restore trust; a failing multi-label role classifier gets replaced by regex in BigQuery replacing ML classifier for dozen CXO roles with much higher recall; an advisor search uses a tiny train word2vec in 15s model and cosine lookup for instant suggestions; a Chrome extension extracts slide financials into Google Sheets after trying LayoutLM LAVa and GPT‑4 Turbo with Nemo guardrails to reduce hallucination.
AI-written summary · about
chapters (11)
- 0:00Credibility in machine learning
- 3:12Why credibility matters
- 7:44UX designers and trust
- 13:06Fusing data sources
- 16:56Simplifying models with regex
- 19:23Fast keyword advisor search
- 22:35Extracting data from documents
- 25:49Using large language models
- 30:45Educating users on AI tools
- 31:05User-centric AI
- 33:53Simple solutions vs AI
related talks
56mDehumanizing Agents: Why Explainability is Crucial in the LLM Era
16mLeveraging ML to Create a Reliable Anti-Fraud System
46mKeynote: How Can We Trust Machine Learning?
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI
54mKeynote: Can you trust your (large language) model?
30mHow to Lie with AI: Understanding Bias, Ethics, and the Hidden Risks in ML
01h 04mCan you trust your (large language) model?
56mData, AI, Hype and Fear - Recipe for Disaster?