Here is the failure this fixes, and it is a quiet one.
You bring the model a genuinely hard question. Which architecture. Whether to take the offer. How to handle the person who keeps doing the thing. It reasons at length, carefully, and it hands you one answer, argued from top to bottom in the same steady voice it uses for everything else.
The answer is usually decent. That is exactly the problem. Nothing about it tells you what it beat, because nothing was ever generated for it to beat. The model committed to a direction in its first few tokens and spent the rest of the response justifying that direction, which is a very good way to make a mediocre idea look inevitable.
Chain of Thought did not fix this. It made the single path visible and inspectable, which was a real advance, and it is still a single path. Depth without width. A carefully reasoned route to wherever the first sentence happened to point.
That is not how anyone competent solves a hard problem. Experts generate several routes, push each one far enough to see where it breaks, and then throw most of them away.
Chain of Thought goes deeper. Tree of Thought goes wider first, then deeper, then deliberately discards most of what it just built.
Yao and colleagues at Princeton published the technique in 2023, and the shape of it is simple enough to describe in a sentence: generate multiple distinct reasoning paths, carry each one to a conclusion, evaluate them against each other, and select. The branching is the visible part. The evaluation and the discarding are where the value actually is.
What it is worth using on
Strategic planning, architecture decisions, negotiation preparation, hiring calls, creative problem solving. The common thread is that the first idea is rarely the best one and the decision is expensive to reverse.
The trade-off, stated honestly
It is verbose. Three reasoned paths plus a comparison costs several times what a single answer costs, in tokens, in latency and in your reading time. That is a real price and it is not always worth paying. Reserve it for the decisions you would regret rushing. For everything else, one good path is enough.