synthetic learning problems

The paper examines in-context learning (ICL) in large language models based on the transformer architecture. It explores how transformers learn in complex scenarios, specifically studying learning with representations. The authors…