LWiAI Podcast #246 - Gemini 3.5 + Omni, Musk Loses, OpenAI vs Erdős
Summary
Hosts Andrei Kerenkov and Jeremy Harris analyze Google I/O announcements, including Gemini 3.5, the agentic Spark assistant, and multimodal Omni models. They discuss the conclusion of the OpenAI antitrust trial, where Musk’s defense failed, and OpenAI’s breakthroughs in mathematical reasoning. The episode also covers Cursor’s new Composer 2.5, rumors of an XAI acquisition, and XAI’s struggling launch of the GrokBuild coding agent amidst significant talent loss.
Topics discussed
Intro and episode overview
Google I/O: Gemini 2.0, Spark, and MCP
Gemini Omni and video generation capabilities
Google Cloud Code, Cursor, and self-improvement
Google's AI tools for scientific research
Genie 3 world models and Waymo simulation
XAI, Cursor, and the Colossus cluster
Elon Musk vs. OpenAI lawsuit verdict
Anthropic valuation and profitability
Andrej Karpathy joins Anthropic
OpenAI product unification and Apple tensions
Cerebras IPO and AI bubble discussion
AI in mathematics and negation neglect
Mechanistic interpretability and NanoGPT speedrun
Claude vs. Codex and terminal agent benchmarks
Deepfake laws and AI self-replication risks
Cybersecurity doubling times and alignment
Synthetic media, AI dramas, and outro
Listen ad-free on Castria