LWiAI Podcast #246 - Gemini 3.5 + Omni, Musk Loses, OpenAI vs Erdős

The 246th episode of the Last Week in AI podcast, hosted by Andrey Kurenkov and Jeremie Harris and recorded on 05/22/2026, summarized and discussed the week's AI news.

Google I/O highlights included Gemini 3.5, with 3.5 Flash emphasized for speed and benchmarks; the always-on agent Gemini Spark running on Google Cloud with MCP tool support; and Gemini Omni for multimodal video generation and editing. Other updates covered were Anti-Gravity 2.0, Gemini for Science, and the Genie world model's navigation using Street View and Waymo simulation.

Coding-agent competition accelerated with Cursor Composer 2.5, described as fine-tuned on Moonshot's Kimi K2.5, and xAI's early Grok Build release. The episode also covered discussion of potential Cursor–xAI ties and xAI's talent churn and compute utilization concerns.

Business and legal items included Elon Musk losing his OpenAI lawsuit on statute-of-limitations grounds; reported OpenAI–Apple partnership tensions; Anthropic agreeing to a $30B funding round at a $900B valuation and projecting its first profitable quarter; and Cerebras' IPO surging about 90%.

Research and safety stories covered OpenAI's result on an 80-year-old Erdős geometry problem, findings on "negation neglect" in training, interpretability work showing multiple redundant circuits per capability, agent benchmarks such as Terminal World, new deepfake takedown enforcement under the Take It Down Act, demonstrations of autonomous hacking and self-replication, rapidly improving AI cyber capabilities, and steps toward image provenance metadata and watermarks.

The episode's timestamped segments also referenced OpenAI co-founder Andrej Karpathy joining Anthropic's pre-training team, Greg Brockman officially taking control of OpenAI's products, autonomous AI research for a nanogpt speedrun, an "All Circuits Lead to Rome" interpretability paper, and Chinese short dramas as AI content machines.