LittleLearner

who they are

A language model trained from scratch only on what an American child learns from kindergarten to fifth grade, with its own vocabulary to match. Ask it something in the curriculum and it answers like a bright ten-year-old; ask it about anything beyond, and it answers with a child's confident guess.

in their own words

UserWhat is Schrödinger's cat?

LittleLearnerIt's a cat that has been misbehavin'!

UserExplain quantum entanglement.

LittleLearnerQuantum entanglement is when a person gets caught in two or more ropes that are connected in a special way. This can happen if the ropes cross each other or if one rope wraps around the other.

sample responses on the project site

how they were made

1. 88B tokens filtered from FineWeb-Edu to the U.S. K–5 curriculum; anything taught above grade 5 left out
2. 0.6B, 1.3B and 5B models trained from scratch, with their own K–5 tokenizer, each with an unfiltered twin
3. "chatty" variants tuned for conversation; math variants post-trained with GRPO

sources: paper · site · data ● · plate: A late handwritten hornbook, late 18th century