AIHacker Newsabout 2 hours ago
LLM trained only on K-5 curriculum
1 min read
Researchers trained LittleLearner models (0.6B, 1.3B, 5B) from scratch on an 88B-token corpus filtered to U.S. elementary-school content. Scaling, GRPO post-training, and in-context learning amplified in-scope K-5 skills but none meaningfully improved out-of-scope performance, indicating pretraining data sets the capability ceiling.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
