Keynote: AI without the BS, for humans
summary
Argues that small models often beat cloud hogging for real users and demos a working pipeline: a Python/Node proxy that extracts token log‑probs and a token heatmap UI to inspect alternatives, a running 3B Llama on a Snapdragon NPU in airplane mode (CPU <10%, fan barely spins), and local OLAMA/DeepSeek 8B experiments plus a smart‑paste form demo — some ports and auth bits wobble but the on‑device, low‑power setups run and return usable summarization and RAG.
AI-written summary · about
chapters (21)
- 0:00Opening remarks
- 1:52Microsoft’s AI push and hype
- 4:15AI perceptions in South Africa
- 6:01AI as stochastic parrot
- 8:26Myths about AI replacing skills
- 10:53Curiosity in AI use
- 14:46AI’s probabilistic nature
- 21:45AI bias and context
- 23:10AI as brainstorming partner
- 24:11Self-referential AI dangers
- 27:35Grounding AI in facts
- 30:45Ethical AI use
- 34:58Global South and AI
- 37:06Case for smaller local AI
- 41:04Live demo: small offline AI
- 44:35Language bias in AI
- 47:10Ethical data for minority languages
- 50:03DeepSeek’s training approach
- 51:49AI reasoning analogy
- 55:24Local AI demos
- 58:15Closing thoughts
related talks
33mKeynote: Boring Wins in AI Agents 🔥 Hard-Earned Lessons
22mSponsored Demo: Cut the Bill, Not the Brains. AI Coding Agents That Ship More and...
54mKeynote: AI-Powered App Development
39mKeynote: Solving the problems of AI in production
15mKeynote: From Platforms to AI Factories. Has Kubernetes Solved It?
59mKeynote: The past, present, and future of AI for application developers
37mEmpowering Developers with local AI
10mKeynote: The Last Mile Problem: Why AI W...