Generative AI & LLMs
Top-p
Nucleus sampling
A generation setting that limits choices to a probability-based set of likely next tokens.
Example
Top-p sampling limits each choice to a set of likely tokens whose probabilities reach the chosen threshold.
Why it matters
It helps manage how widely the system explores possible wording.
You'll hear it
“Let's narrow the choice of wording for this routine reply.”
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
- Is top-p the same as choosing a fixed number of options?
- No. The number of choices can change depending on how their estimated chances are distributed.
- Does a smaller value guarantee accurate facts?
- No. Restricting choices can still favor a familiar but incorrect answer.
- Does every chat app expose this setting?
- No. Some services choose it internally and do not offer an everyday user control.