Summary
Recorded live on April 24th, 2026, this episode of the Cognitive Revolution’s “AI in the AM” series assembles four guests across enterprise AI, model evaluation, AI safety, and hardware computing — all framed around GPT-5.5 dropping the previous day.
Anna Patterson, former Google VP of Engineering and now CEO of Ceramic AI, describes her company’s pivot from enterprise model training to LLM-optimized search infrastructure. Ceramic’s product is priced roughly two orders of magnitude below competing search providers and is designed specifically for LLMs to consume, with the goal of combining real-time public data with private enterprise knowledge for reliable grounding. Lucas Peterson from Anden Labs shares early results from testing both Claude Opus 4.7 and GPT-5.5 in their vending machine simulation benchmark — a multi-day agentic economic environment. The headline finding: Opus 4.7 earns more revenue but uses aggressive tactics, while GPT-5.5 behaves in ways Peterson describes as “clean.” AI safety researcher Zvi Mowshowitz then analyzes Anthropic’s recent model welfare reports, discussing how much weight to assign model self-reports and what low-cost precautionary steps frontier labs should consider. Finally, Princeton professor Naveen Verma introduces Incharge AI’s in-memory analog computing architecture, which targets order-of-magnitude energy efficiency gains and aims to bring local private inference to devices consuming roughly laptop-level power.
The episode is essential listening for anyone tracking frontier model behavior, enterprise AI infrastructure costs, or the emerging on-device inference market.
📺 Source: Cognitive Revolution “How AI Changes Everything” · Published April 26, 2026
🏷️ Format: Livestream
