Mixture of Experts OpenAI social network, Anthropic’s reasoning study and humanoid half-marathon
8 snips
Apr 25, 2025 
Join Gabe Goodhart, Chief Architect at AI Open Innovation, Kate Soule, Director at Granite, and Marina Danilevsky, Senior Research Scientist, as they dive into the fascinating intersection of AI and social media. They discuss the rumors of OpenAI's potential social network and ponder its implications. The guests also explore Anthropic's study on AI reasoning, highlighting its complexities. Finally, they share insights on a humanoid robot half-marathon, questioning whether robots might be the next race contenders in our future.
AI Snips
Chapters
Transcript
Episode notes
OpenAI Social Network Data Play
- OpenAI's rumored social network is likely a data collection play to improve conversational models.
- The platform may attract a specific slice of creativity like humor and memes that synthetic data misses.
Chain-of-Thought is 'Warming Up'
- Chain-of-thought reasoning in models is more of a probabilistic output booster than true explainability.
- The model's reasoning traces don't faithfully represent internal decision processes but aid human understanding.
Partner to Protect Wikipedia Data
- AI companies should partner with Wikipedia to reduce harmful scraping and support infrastructure costs.
- Building official datasets and hosting mirrors can help preserve open knowledge and sustain Wikipedia.