#216 - Grok 4, Project Rainier, Kimi K2
Summary
This episode analyzes xAI’s Grok 4 launch, which leads frontier benchmarks but sparked controversy over anti-Semitic outputs and alignment with Elon Musk’s views. The discussion covers the rise of AI-powered browsers from Perplexity and OpenAI challenging Google Chrome, alongside updates to coding agents like Cursor and Replit. Finally, it highlights major business developments, including Lovable’s $2 billion valuation and Amazon’s massive Project Rainier supercluster for Anthropic.
Topics discussed
Sponsors: OutShift, Box, and 988 Lifeline
Intro and episode overview
Grok 3 and 4 performance, RL scaling, and benchmarks
Grok alignment issues, political bias, and safety concerns
Perplexity Comet, AI browsers, and Cursor pricing changes
Anthropic Project Rainier and Amazon Trainium chips
OpenAI leadership changes and stock compensation
New model releases: Qwen, Fish Audio, and DeepSeek
Research: Math reasoning transfer and RL vs SFT
Meta study on AI impact on developer productivity
Safety research: Goal misgeneralization and model agreement
DeepMind stealth evals and alignment faking in Claude
App store review manipulation, EU regulations, and virology benchmarks
Listen ad-free on Castria