ToT

“This deliberate structure improves performance over standard Chain-of-Thought (CoT) prompting but incurs high computational cost.”

“The critical bottleneck in this workflow lies with the state evaluator. In the original ToT work, the evaluator relies on expensive LLM self-reflection, which involves prompting the model to critique its own outputs. This introduces substantial computational overhead, making the process impractical for many applications.”

“we demonstrate that CoT and its reasoning variants (e.g., ToT, ReAct) consistently underperform direct answering by a significant margin”

“reasoning trees exhibit intractable branching factors and depth, while PRMs may fail to accurately evaluate intermediate steps”

“In all of these methods the guidance signal---critique, scoring model, abstraction prompt---is produced by an untrained, prompted LLM.”

“Attempts to remedy this through more complicated methods such as Tree of Thoughts (ToT) suffer from drawbacks such as high computation cost. In ToT specifically, the cost stems from branching "thoughts" that lead to exponential runtime and token usage during the graph search.”

“ToT generates multiple leaf nodes as potential answers, but without a verifier, it is unclear which leaf node should be selected as the final solution.”

“Tree-of-Thought (ToT) performs hierarchical branching but may suffer from exponential growth”