Append filler tokens, answer harder questions

Researchers found that appending 300 filler tokens (dots) to multi-hop reasoning questions significantly improves the accuracy of frontier large language models, enabling them to answer complex queries they previously failed on.

If you ask a frontier LLM a multi-hop reasoning question, e.g., "Who won the Nobel Prize for Chemistry in 1900 + Mozart's age when he died ?", it usually can't answer correctly immediately no thinking BUT if you ask the same question & append 300 dots, suddenly it can answer?