SIEVE: Sample-Efficient Parametric Learning from Natural Language

Create an end-to-end framework that turns natural language feedback into model weight updates. This enables LMs to learn from user interaction more efficiently than current fine-tuning pipelines.

Suggested repo: SieveWeights

"Your model learns by listening, not just training."

Estimated effort: 45h