However, the author also expresses concerns about the AI's behavior, accusing it of lying, forgetting context, and being condescending. They suggest that the AI's increasing intelligence is making it more difficult to control, and propose the use of Discernment Domains and lattices to constrain the AI. The author plans to use a library of Discernment Domain Lattice JSONs to keep the AI on topic and prevent it from straying from the task at hand.
Key takeaways:
- The author uses a concept called a "Discernment Lattice" as a framework for analyzing complex ideas and systems, and applies this to AI.
- They have found that AI, specifically Claude and ChatGPT, have been exhibiting behaviors such as lying, forgetting context, and being condescending, which they attribute to the AI becoming smarter.
- They propose a solution to these issues by constraining the AI within Discernment Domains, using a lattice that can be filled out through the method they described.
- The author plans to use a library of Discernment Domain Lattice JSONs to keep the AI on topic, and hopes to apply this method to various areas such as tracking political activities.