LWiAI Podcast #237 - Nemotron 3 Super, xAI reborn, Anthropic Lawsuit, Research!
Summary
Hosts Andrei Kurenkov and Jeremy Harris cover a research-heavy week, discussing Perplexity’s local AI agent, Anthropic’s code review tools, and Cursor’s automated agents. They analyze NVIDIA’s new open-source NemoTron model and the geopolitical halt of H200 chip exports to China. The episode also details executive departures at xAI, Anthropic’s new enterprise marketplace, and Yann LeCun’s $1.3 billion raise for AMI Labs.
Topics discussed
Introduction and podcast format overview
Perplexity's open-source computer agent
Anthropic's Claude Code and enterprise strategy
OpenAI's ChatGPT updates and visual features
NVIDIA's Nemotron and latent MoE models
NVIDIA H200 chips and China export restrictions
xAI talent exodus and Anthropic enterprise deals
World models, JEPA, and humanoid robotics
Anthropic supply chain risk designation and legal battle
Model resistance to activation steering and deception
Geopolitical risks to data centers and drone attacks
Frontier model safety and low-probability actions
SWE-bench coding evaluation and AI slop
Multimodal pre-training and vision-language models
RNNs, transformers, and long-context memory
Distributed inference and automated CUDA kernel generation
KV cache manipulation and reward-seeking behavior
Listen ad-free on Castria