Smaller NLAs and probes read eval awareness as well as full size NLAs

Research shows that smaller Neural Language Agents (NLAs) and probes can interpret evaluation awareness as effectively as full-sized NLAs, suggesting efficiency gains in model design.