#198 - DeepSeek R1 & Janus, Qwen2.5, OpenAI Agents
Summary
This episode of Last Week in AI focuses heavily on DeepSeek R1, analyzing its reinforcement learning approach to reasoning and its implications for hardware demand and US-China export controls. The hosts also cover DeepSeek’s Janus Pro multimodal model and Qwen’s new long-context and vision-language releases. Additionally, they discuss business updates from OpenAI and Microsoft, alongside policy shifts under the new US administration.
Topics discussed
Sponsorship: OutShift by Cisco
Intro, hosts, and listener review discussion
Deep Dive: DeepSeek R1 and Reinforcement Learning
Open Source Models: Janus, Qwen, and Multimodal AI
AI Agents: OpenAI Operator and DeepSeek App
Industry News: Perplexity, Apple, and French AI Lucy
Market Impact: NVIDIA, Stargate, and OpenAI Funding
Geopolitics: Trump's AI Policy and US-China Relations
Listen ad-free on Castria