Fireside Chat
summary
Fireside Chat summarizes concrete claims and tools: large models improve with self-supervised learning so one pretrained model serves many tasks; open research and toolchain contributions like PyTorch, detectron2 and fairseq accelerate the field. Infrastructure limits matter — brains use ~25 watts while current AI is ~100,000× less efficient, so specialized hardware and quantization are essential. FAIR funds bottom-up 3–5–10 year projects, including multilingual content moderation and an “open category” chemistry/protein effort aiming to design useful compounds.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:00Yanns journey with AI
- 2:40Examples of AI research projects
- 4:40bottlenecks in AI research
- 6:51selfsupervised learning
- 8:51sustainability of AI
- 12:02new paradigms
- 15:22early interest in intelligence
- 19:03challenges of AI
- 21:39support for open science
- 26:32progress in NLP
- 28:42future of AI
- 31:36conclusion
