AI Safety & Responsible AI
Model alignment
Techniques used to shape a model so its behavior better follows intended preferences and rules.
Example
An assistant consistently follows the organization's rules even when users pressure it not to.
Why it matters
It helps make learned abilities usable within intended rules and preferences.
You'll hear it
“It follows the rules in normal cases; what happens when instructions conflict?”
Can you control it?
Sometimes
Sometimes. Some products or workflows let you influence model alignment, while others keep it fixed or manage it automatically. Check the settings and documentation for the tool you are using.
Common questions
- Does alignment work erase every unwanted behavior?
- No. Improvements can leave gaps or behave differently in unfamiliar situations.
- Can shaping behavior affect usefulness?
- Yes. Some restrictions or preferences can change what the system can helpfully do.
- Is the model the only part that matters?
- No. Surrounding tools, permissions and application rules also shape the final behavior.