Rethinking Pre-Training for Agentic AI [Aakanksha Chowdhery] - 759
The TWIML AI Podcast with Sam Charrington · 51:54
Aakanksha Chowdhery, who trained PaLM and Gemini 1.0/1.5 at Google and now works at Reflection, argues that reliable agents need changes to pre-training itself, not just more post-training on top of models built for c...