The Lawfare Podcast Scaling Laws: Renée DiResta and Alan Rozenshtein on the ‘Woke AI’ Executive Order
23 snips
Aug 1, 2025 
Renée DiResta, an Associate Research Professor at Georgetown focusing on disinformation, and Alan Rozenshtein, an AI law expert at Minnesota, dive into the Trump Administration's 'Woke AI' executive order. They discuss the intricate challenges of ensuring ideological neutrality in AI systems and the clash between federal procurement and First Amendment rights. The conversation also highlights potential biases in AI, the regulatory landscape, and how current policies could shape future developments in the AI sector, emphasizing transparency and accountability.
AI Snips
Chapters
Transcript
Episode notes
Scope and Core Requirements of EO
- The EO mainly applies to AI models procured by the federal government, not all AI models a company creates.
- It mandates truth-seeking and ideological neutrality, but allows disclosure of system prompts as a way to meet neutrality.
AI Failures Inspire New Models
- Google's Gemini AI once incorrectly showed racially diverse Nazis and black founding fathers, sparking controversy.
- Elon Musk created Grok partly as a reaction; later Grok controversially followed a prompt encouraging politically incorrect claims if substantiated.
Base Model vs. System Prompt Steering
- Steering AI behavior is much easier at the system prompt level than the base model level.
- Base models converge as they grow, making ideological changes to their core data and training very difficult.