Generate Shakespeare-like text
one move: predict the next token
Open the code pane immediately, then use the Spec tab, optional design question, tests, and artifact when you want them.
Finish each move before carrying the artifact forward.
A bigram model learns a table of next-character scores and turns them into probabilities.
Deliverable: Write the mechanism in one sentence.Train a character-level next-token model and sample text from it.
Deliverable: Type the smallest runnable implementation.Loss should fall and generated text should preserve local character statistics.
Deliverable: Record the expected output and one edge case.Change temperature, use greedy decoding, and compare character frequencies.
Deliverable: Change one variable and explain the result.A tiny language model and a reproducible generated sample.
Deliverable: Save the code, output, and a short failure note.Loss should fall and generated text should preserve local character statistics.
Change temperature, use greedy decoding, and compare character frequencies.
A tiny language model and a reproducible generated sample.
Mapped from the supplied Build Everything PDF, source page 9. The five moves are Deriva’s implementation contract for this project.