Keynote
summary
Argues that transfer learning fine‑tuning is the core breakthrough for language models and that ULMFiT, built in a tiny Jupyter notebook on PyTorch, beats prior state of the art on the IMDb long‑document benchmark. Demonstrates that the three‑stage fine‑tuning pipeline underpins later GPT-style success, warns that heavy use of agents produces more code but less shipped product and erodes human craft, and describes a hands‑on solve it workflow and new fast.ai course that keeps humans in the loop.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:26First large language model
- 4:59AI agents and outsourcing
- 9:56AI and human craftsmanship
- 13:03Open-sourcing frontier AI models
- 16:14Democratic AI compute access
- 17:37China and pioneering companies
- 18:59Future of human-centered AI
related talks
19mIntro to Fine-tuning
54mKeynote: Can you trust your (large language) model?
21mBeat GPT 4 and Build the Best AI For You: Mobilizing First-Party Data with Fine-Tuning
21mBeat GPT 4 and Build the Best AI For You: Mobilizing First-Party Data with Fine-Tuning
33mBuild Better AI with First-Party Data and Fine-Tuning
27mEmbeddings, Transfer Learning and More: LLM Tailored to You
54mKeynote: Can you trust your (large language) model?
44mHow to build your own Tiny Language Model from scratch