Let's Think Dot by Dot: Hidden Computation in Transformer Language Models

arXiv:2404.15758v1 Announce Type: new
Abstract: Chain-of-thought responses from language models improve performance across most benchmarks. However, it remains unclear to what extent these performance gains can be attributed to human-like task decomposition or simply the greater computation that additional tokens allow. We show that transformers can use meaningless filler tokens (e.g., '......') in place of a chain of thought to solve two hard algorithmic tasks they could not solve when responding without intermediate tokens. However, we find empirically that learning …

abstract arxiv benchmarks computation cs.ai cs.cl hidden however human human-like language language models performance responses show think thought tokens transformer transformer language models transformers type