Rohan Anil Calls Neuralese Agent Communication Pretty Bad

Rohan Anil stated that agents communicating in neurelese is pretty bad. The post came from the researcher who co-founded CoreAutoAI after serving as Distinguished Engineer at Google DeepMind on Gemini pretraining. Neuralese refers to internal neural network representations that agents use instead of human language. A generated headline summarized the remark as calling such communication ineffective. The source summary noted the concern the comment highlights about agents relying on these representations for interaction. The statement stands as the direct claim from the visible post.