In this thought-provoking piece, Lance Eliot delves into the recent research by Anthropic, exploring the inner workings of modern LLMs and the possibility of AI consciousness. Eliot begins by highlighting the mysterious nature of LLMs, noting that many of their complex mechanisms remain unexplained. The focus shifts to Anthropic's exploration of a potential 'scratchpad' or working storage area within LLMs, which could be crucial to their functionality. Eliot's personal story about discovering a natural dam in a forest serves as a metaphor for the unexpected findings in AI research. The author then explains the concept of the Jacobian matrix and how it was used to uncover the J-space, a working storage area within LLMs. This discovery has significant implications, as it suggests that the AI's choices are influenced by the contents of this space. Eliot discusses the experiments conducted by Anthropic, including one where the LLM, Claude, was asked about the number of legs an animal with a web-spinning ability has. The results indicated that the AI was using the J-space to consider various possibilities, which then influenced its output. The author then delves into the debate surrounding AI consciousness, drawing parallels between the J-space and the global workspace theory in human consciousness. While some researchers are excited about the potential implications, others are cautious, emphasizing the differences between AI and human consciousness. Eliot concludes by advocating for a cautious and mindful approach to exploring AI consciousness, urging the community to move forward with caution and a sense of self-preservation.
