Last Week in AI Last Week in AI

#231 - Claude Cowork, Anthropic $10B, Deep Delta Learning

Jan 21, 2026

Summary

This episode covers Anthropic’s new Co-work agent, Google’s Gemini updates, and major funding rounds for Anthropic and xAI. It also discusses NVIDIA’s H200 chip supply constraints, OpenAI’s Cerebras compute deal, and CoreWeave’s financial adjustments. Finally, it highlights research on scaling reinforcement learning to prevent catastrophic forgetting in AI models.

Topics discussed

Sponsors: OutShift, Box, and Facials Intro and show format update Anthropic Claude 4.5 and Computer Use features Google Gemini updates and search competition Google Gmail AI features Salesforce AI and Slack integration Anthropic and xAI funding rounds NVIDIA H200 chips and China export controls Cerebras IPO and CoreWeave debt restructuring LM Arena funding and commercialization Cascade RL and DeepSeek research papers DeepSeek architecture and residual stream analysis Hybrid Mamba-Transformer models Scaling laws and neural network theory Retrieval augmented generation and memory Positional embeddings and rotary clocks Anthropic safety and jailbreak detection China AI gap and US export control politics Outro and closing remarks
Listen ad-free on Castria

Episode summaries and topics are generated from the audio by Castria — a free podcast app that skips ads automatically.