Model Training & Adaptation
Reinforcement learning from AI feedback
RLAIF
Training that uses AI-generated preference or critique signals instead of only human feedback.
Example
A stronger model critiques candidate answers that are then used to improve another model.
Why it matters
It can provide feedback at a scale that would be costly for people to review alone.
You'll hear it
“The automated reviewer missed the same mistake as the writer.”
Can you control it?
Set by the team
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Can feedback repeat another AI's bias?
- Yes. Mistaken judgments or preferences can be passed into the system being trained.
- Does this eliminate the need for human judgment?
- No. People still need to choose goals and check whether the feedback is useful.
- Can an AI judge style better than factual truth?
- It may find surface qualities easier to assess than claims requiring reliable outside evidence.