LittleLearner：一个只学美国小学课程（K–5）的 LLM 沙盒，研究模型能力边界。

Talk to LittleLearner

The hosted 5B model, live in your browser.
Open in a new tab ↗
if the chat doesn’t load below.

ℹ️ Research artifact: This model is not designed or safety-aligned for children. It is intended for academic research purposes.

A controlled sandbox for studying how models acquire knowledge

Modern LMs are trained on everything at once, so it is hard to tell whether a new skill
was learned or merely elicited. We constrain the training distribution itself: an 88B-token
corpus filtered to the U.S. elementary-school curriculum, with models trained from scratch on it and
matched unfiltered controls.

Dataset

LittleCurriculum

An 88B-token corpus distilled from FineWeb-Edu through a five-stage filtering pipeline aligned with Common Core standards (K–5).
Concepts, facts, and vocabulary taught above Grade 5 are explicitly excluded.

Models

LittleLearner

Three scales (0.6B / 1.3B / 5B) trained from scratch on LittleCurriculum: chattable models
with an interpretable knowledge boundary. Each ships with a matched Unfiltered control
for clean comparison.

Findings

Elicitation, not acquisition

In our experiments, scaling, SFT+GRPO post-training, and in-context learning amplify what the
curriculum taught, but none meaningfully improves out-of-scope performance, indicating that the
pretraining filter sets the effective capability ceiling.

Model checkpoints

LittleLearner at three scales (0.6B / 1.3B / 5B), each with a matched Unfiltered control
sharing its architecture, tokens, and recipe.

Base: the pretrained model.

GRPO: math specialists post-trained on MathCAMPS; responses may exhibit a tendency toward math-oriented output.

Chatty: variants tuned for general chat behavior.

Scale
LittleLearner · K–5
chatty
Matched control · unfiltered

Capability stays inside the curriculum

Can standard interventions push a model past what its pretraining data taught it?
With the boundary under experimental control, we can ask cleanly. In our experiments, each
intervention amplifies in-scope ability; none of them meaningfully improves out-of-scope
performance.

Scaling
Post-training
In-context learning

Scaling

Scaling model size improves performance within the model’s controlled knowledge exposure and
extends modestly to problems along the same learning trajectory, but yields little improvement on
problems requiring more advanced capabilities outside the exposure.

MathCAMPS accuracy by grade, across model size

Post-training

Post-training through GRPO significantly boosts in-scope K–5 capabilities, but fails to
recover out-of-scope beyond-K–5 capabilities, even when training with out-of-scope data.

Post-training amplifies K–5, not the beyond-K–5 gap

In-context learning

In-context learning with the prompts we test does not unlock new reasoning capabilities in
beyond-K–5 for our trained 5B LittleLearner.

Accuracy by prompting condition

What will you teach it?

Because LittleLearner’s training exposure is explicitly specified, behavioral and
representational changes can be related directly to the concepts you introduce. Three directions
we’re excited about:

01
RL & discovery

Can RL create capability?

The prior is restricted to K–5, so capabilities that emerge under RL can be attributed to
the RL process itself. A tractable proxy for reward-driven discovery.

02
Continual learning

Watch a concept being learned

Introduce negative numbers and measure sample efficiency, retention, and interference. Or
probe behavior near the boundary: does it answer, abstain, or hallucinate?

03
Educational science

Machine vs. child learners

Specified exposure enables controlled human-model comparison. Do models and children need
similar exposure to learn fractions, or make similar errors on word problems?

＋
Your turn

Bring your own question

A known boundary turns your idea into a clean experiment!

If you find this work useful

Please cite our paper:

Copy@misc{littlelearner2026,
title={LittleLearner: Language Models Under Pedagogically-Controlled Knowledge Exposure},
author={Fanfei Li and Jana Zeller and Manuel Prada-Corral and Thaddäus Wiedemer and Prasanna Mayilvahanan and Ryan Cotterell and Wieland Brendel},
year={2026},
eprint={2608.13545},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2608.13545}
}