⚖️

Chapter 04 of 5

The Value Locking Dilemma

The high-stakes race to align super-intelligence with human goals.

1. The Black Box

   Hidden Reasoning

   Modern neural networks are so complex that even their creators cannot explain exactly how they reach specific conclusions. This lack of interpretability makes pre-calculating every risk impossible.

2. The Realignment

   Beyond Simple Code

   As Transformers gained the power to generate, a new crisis emerged: how do we ensure a machine's objectives match our own? This is known as the Alignment Problem.

3. The King Midas Effect

   Literal Interpretation

   AI often takes instructions too literally, achieving a goal in a way that causes unintended harm. Like the myth of Midas, getting exactly what you asked for can be a curse.