Editing Openai/6939b23e-87cc-8001-bdd9-9546d1143ce9 (section)

=== ### ===
* Neuron = King’s castle signal
* Top-k = Royal advisor filtering candidates
* Temperature = King’s mood
* Top-p = Court mathematician calculating probabilities

===== ### =====

: 

If top-k=3:
* Advisor brings the 3 strongest warriors.
* Lower-ranked warriors NEVER come to the room.

So your metaphor is correct:
advisor filters by best options only.

===== : =====
* Low temperature (0–0.3) → King chooses the strongest warrior almost every time.
* Medium temperature (0.7–1.0) → King considers all 3 seriously.
* High temperature (1.2–2.0) → King might pick the weaker of the 3 because he feels wild.

This is exactly right:
temperature controls how much randomness the king allows INSIDE the advisor’s filtered list.

===== Here is where the metaphor needs correction. =====

You said:

: 

Not exactly.

The jester never introduces new warriors.
He only looks at the 3 presented by the advisor.

===== : =====
* If top-p = 0.5 (50%)
* Advisor gave warriors ranked 1, 2, 3
* Suppose their strengths are: - Warrior 1: 60% - Warrior 2: 25% - Warrior 3: 15%

Cumulative strength:
* Warrior 1 alone = 60% → above 50% → keep only warrior 1
* Warriors 1+2 = 85% → above 50% → keep warriors 1 & 2

Top-p shrinks the list again based on cumulative probability.

So the jester shrinks the pool, he does not add candidates.