Anthropic has introduced a new feature that will allow the chatbot to 'dream' and fix its 'memories' while it does so. "A dream reads an existing memory store alongside past session transcripts, then produces a new, reorganised memory store," Anthropic said. Users can also see in real time what Claude is writing and reading while it's dreaming.