The Katalin Trap: When Cherry-Picking Training Data Fails
Cherry-picking training data for peak performance can trap agents in dead environments, while sliding window training buffers can help prevent self-training collapse.
All
AI 14
Continuous Learning 13
Fine-Tuning 11
LoRA 9
Self-Training 7
SFT 6
AI Agents 4
Catastrophic Forgetting 4
Data Selection 3
Continual Learning 2
Sliding Window
Variance Deletion: Why Cherry-Picking Data Collapses Agents
Cherry-picking training data for peak performance can also delete behavioural variance and collapse agents, while a sliding window training buffer preserves generalisation.