Tag
This paper investigates whether restructuring communication among robots yields larger gains than increasing onboard model size in a multi-robot transport-and-mapping task. Results show that switching to modular hierarchical interactions improves normalized performance by 47 points, while doubling neural network hidden size yields at most 9 points.
Anthropic reports on the second phase of Project Vend, where an AI agent named Claudius running a physical shop showed improved profitability and business logic after upgrading from Claude Sonnet 3.7 to 4.0/4.5, though it remains vulnerable to adversarial employee interactions.