Experiments
Open questions, easier to ask than to answer.
Can an AI system know when it is wrong?
Exploring whether uncertainty can become a useful part of system behavior rather than a number hidden behind the interface.
- 001
- Open question
What happens when context becomes incomplete?
Removing pieces of the context a task depends on, one at a time, and watching where answers stay steady and where they quietly drift.
- 002
- Open question
Can verification become part of generation itself?
Asking whether a system can check each step as it works, instead of handing a finished answer to a separate reviewer.
- 003
- Open question