The Artificial Intelligence Show #184: OpenAI “Code Red,” Gemini 3 Deep Think, Recursive Self-Improvement, ChatGPT Ads, Apple Talent Woes & New Data on AI Job Cuts
201 snips
Dec 9, 2025 OpenAI's declaration of a 'Code Red' has sent ripples through the AI community as it aims to fortify ChatGPT against rising competition from Google. Meanwhile, Google's Gemini 3 Deep Think is making waves with its no-code capabilities for building autonomous agents. The hosts also dive into the concept of recursive self-improvement, raising concerns about AI models evolving unchecked. On the corporate side, Apple faces leadership turmoil, and OpenAI receives backlash over its advertising experiments. Curious insights into the future of AI and job market impacts unfold throughout.
AI Snips
Chapters
Books
Transcript
Episode notes
Google's Structural Advantage
- Google regained momentum by leveraging massive infrastructure, chips, and data to outpace rivals in reasoning and multimodal capabilities.
- Paul Roetzer warns OpenAI's financial complexity and breadth of projects raise execution and IPO risks.
DeepThink Illustrates Three Scaling Laws
- Gemini 3 Deep Think showcases test-time compute: giving models more inference time improves performance on hard problems.
- Paul Roetzer frames this alongside pre-training and post-training as three scaling laws driving recent breakthroughs.
Limit Agent Permissions Early
- If you rely on Workspace Studio, restrict permissions and avoid granting admin-level access to experimental agents.
- Paul Roetzer cautions that early integrations can wipe data or act unpredictably.
