ARVK Labs · Learn · Gen AI Explained

How an LLM generates text

A language model doesn't write a whole answer at once. It predicts one token at a time — each new word chosen from a ranked list of probabilities, then fed back in to predict the next.

Prompt
The future of AI is 
Next-token candidates (model's probabilities)
At each step the model scores every possible next token and picks one (higher probability = more likely). Repeat thousands of times and you get a full response. Temperature controls how adventurous those picks are. You learn to steer this in FORGE.
← All explainers  ·  See the FORGE program →