The latest research in AI language models, including algorithmic progress and multistep consistency models.
A new LLM model called Command-R, designed for large-scale production workloads.
The announcement of Meta's investment in AI infrastructure, including two 24k GPU clusters and plans for continued growth.
A discussion on the potential security risks of model-stealing attacks on language models.
Contact: sergi@earkind.com
Timestamps:
00:34 Introduction
01:42 Command-R: Retrieval Augmented Generation at Production Scale
03:48 Building Meta’s GenAI Infrastructure
05:41 MLX Server
06:37 Fake sponsor
08:45 Algorithmic progress in language models
10:23 Multistep Consistency Models
11:35 Stealing Part of a Production Language Model
13:31 Outro
