Emergent Capabilities Arise Randomly from Learning Sparse Attention Patterns
Provides mechanistic insight into why emergent capabilities appear abruptly in large language models, relevant for understanding and predicting model behavior.
The paper shows that emergent capabilities in transformer language models arise stochastically during training due to the abrupt learning of sparse attention patterns, with larger models acquiring them earlier. On synthetic tasks, scaling attention heads improves learning efficiency, while increasing head dimension yields diminishing returns.
Neural scaling laws for transformer language models predict smooth improvements in pretraining loss with increasing parameters, but downstream capabilities such as in-context learning are known to emerge abruptly past a certain model scale. In this paper, we show that emergent capabilities arise stochastically throughout training, with larger models acquiring them earlier on average. We demonstrate that the emergence of capabilities such as pattern completion and indirect object identification corresponds to the abrupt learning of task-relevant attention patterns. To isolate this phenomenon, we train transformer models on synthetic linear map and cellular automata datasets, and we show that the difficulty of learning attention patterns depends on context length and pattern sparsity. Moreover, scaling the number of attention heads improves learning efficiency on our synthetic tasks, while increasing the head dimension yields diminishing returns past a minimum capacity. We additionally investigate architectures with alternative attention mechanisms, showing that MLP-Mixer outperforms a transformer on linear map tasks with complex attention patterns. Our findings provide a mechanistic insight into emergence, showing that downstream capabilities arise abruptly due to the intrinsic difficulty of learning sparse attention patterns in transformer models.