Anthropic researcher Jack Lindsey tests AI introspection via neural injections

Recent experiments examining whether large language models can introspect have sparked debate among researchers. While Anthropic researcher Jack Lindsey found that models could sometimes detect and describe artificially injected internal "thoughts," follow-up studies by other researchers suggest that models may only be capable of anomaly detection rather than true introspection. Consequently, experts remain divided over whether these responses indicate genuine awareness of internal states or simply stem from simpler mechanisms. Open article →