#237 - Nemotron 3 Super, xAI reborn, Anthropic Lawsuit, Research!!!
Summary
This episode covers Perplexity’s local AI agent, Anthropic’s code review tools and enterprise marketplace, and Cursor’s automated coding agents. It also discusses NVIDIA’s new open-source model and halted H200 chip production for China. Finally, the hosts analyze XAI’s leadership exodus and Yann LeCun’s $1.3 billion raise for world models.
Topics discussed
Sponsors: OutShift and Box Agent
Apple Intelligence and OpenClaw security
Cloud Code update and automated PR reviews
NVIDIA NemoTron, MoE architecture, and quantization
US-China GPU export controls and HBM memory
xAI leadership changes and refounding
Anthropic Cloud Marketplace and billing
European AI funding and JEPA architecture
Anthropic supply chain risk designation and legal battle
Activation steering resistance and model self-correction
Data center security and warfare implications
Inference compute budgets and simulated randomness
SWE-bench limitations and code review quality
Multimodal pre-training and vision-text scaling
Memory caching and long-context recall
Untied Ulysses: Memory efficient context parallelism
Agentic RL for high-performance CUDA kernel generation
Models detecting prior concept injections
Reward-seeking behavior and alignment theory
Listen ad-free on Castria