The cat sat on themat
A model rolls loaded dice for every single word. You load the dice.
The spread here is broad but sensible. Turn the heat up and the unlikely picks come alive.
The cat sat on the·····
A clear favourite dominates, but 10 words are in the running. The uncertainty in the field is 2.6 bits.
- mat42%
- floor15%
- sofa11%
- rug9%
- bed7%
- roof6%
- chair4%
- step3%
- fence2%
- windowsill1%
The word probabilities are illustrative and approximate. Provided by the Institute of AI for interest and learning.
Three settings behind
every reply.
Temperature
Temperature divides the scores before they become probabilities. Low values sharpen the model onto its favourite token; high values flatten the field so surprises can win.
Top-p (nucleus)
Top-p keeps only the smallest set of tokens whose probability adds up to at least p, then samples from those. It is a soft cap that adapts to how confident the model is.
Top-k
Top-k is a hard cap: only the k most likely tokens are ever considered. It is simple and blunt, and often combined with top-p to trim the long tail.
Greedy vs sampling
Greedy decoding always takes the single most likely token, giving repeatable but flat text. Sampling trades some reliability for variety and creativity.
Membership is free. Accreditation is the standard.
Keep
exploring.
Get AI is a collection of games, experiences, and tools that make AI easier to understand from the Institute of AI.