The article argues that current large language models lack a critical capacity called "situation perception," which is essential for achieving artificial superintelligence. This missing ability involves constructing and acting within internal simulations of possible worlds across latent time.
- Situation perception requires abstract prediction, long-term compressed memory, and active learning guided by objectives.
- The authors analyze why modern large language models remain incomplete in this regard.
- The work proposes appropriate tests for measuring progress toward machines that can simulate futures and pursue self-directed goals.
The authors consider this important because they believe the path to artificial superintelligence depends on developing these capabilities, which would allow machines to possibly judge their own creators.