Keynote: The past, present, and future of AI for application developers
summary
Walks through Eliza and Markov generators, then trained GPT‑2 in C on NDC session texts for 500 iterations on a laptop (loss drops ~12→7 and outputs barely‑coherent phrases). Explains transformer scaling (GPT‑1 ≈100M, modern models hundreds of billions), shows a .NET RAG chat over PDFs with citation links and real‑time voice+function calling to call web APIs, a browser demo running a 300MB DeBERTa v3 zero‑shot classifier, and a live car‑listing assistant that edits JSON and works.
AI-written summary · about
chapters (13)
- 0:00AI for developers intro
- 3:18AI origins in 1950s
- 8:20Markov to language models
- 14:37Transformer breakthrough
- 16:22Training GPT2 in real time
- 21:43Token prediction to chat
- 29:04AI chat interfaces today
- 35:23AI for business workflows
- 40:38Small language models
- 44:28Reasoning models limits
- 47:39AI UI enhancements
- 50:05Voice assistants with AI
- 58:14AI future summary
related talks
58mKeynote: The past, present, and future of AI for application developers
01h 01mKeynote: Transformers: The Rise of ChatGPT
45mBringing the Power of AI to Your Application
19mLet's Talk GenAI
45mKeynote: Django Reimagined For The Age of AI with Marlene Mhangami
55mEveryday AI: Practical Applications Beyond the Chatbot
34m"The Past, Present, and Future of GenAI"
43mKeynote: Applied NLP in the age of Generative AI