ChatGPT-4 Prompt Engineering: The Tree of Thoughts Method - WOW!
Based on All About AI's video on YouTube. If you like this content, support the original creators by watching, liking and subscribing to their content.
Tree of Thoughts prompting improves problem-solving by generating multiple candidate reasoning paths and allowing backtracking instead of committing to a single approach.
Briefing
Tree of Thoughts prompting is presented as a way to make large language models solve problems more reliably by exploring multiple reasoning paths instead of committing to a single line of thinking. Rather than “walking one trail and hoping,” the method generates several candidate approaches, evaluates them against criteria, expands each option into realistic scenarios, and then ranks the best path forward. The practical payoff is that the model can backtrack from dead ends and choose among alternatives with more deliberate selection—an approach framed as improving problem-solving performance compared with simpler chain-style prompting.
The walkthrough breaks the process into four phases. First comes a “brainstorm” phase that asks for three or more distinct solutions to a user’s problem, encouraging variety across factors. In the example, a user wants to ask a boss for a raise while describing themselves as shy, conflict-avoidant, and employed for six years, with a baby on the way. The prompt steers the model to generate options using background, timing, preparation, professionalism, and value proposition. The resulting solutions include: (1) a written request via email or formal letter to reduce interpersonal pressure, (2) leveraging the performance review cycle to discuss compensation in a more expected, less confrontational setting, and (3) a direct conversation to schedule a meeting—straightforward but potentially stressful.
Next is an “evaluation” phase that scores each option by pros and cons, initial effort, implementation difficulty, potential challenges, and expected outcomes. The model assigns probability-of-success estimates and confidence levels. In the example, the performance review route is treated as naturally aligned with company norms and timing, the written request as lower-pressure but slower and potentially less urgent, and the direct conversation as effective but higher anxiety and higher risk of rejection. The evaluation also surfaces practical constraints like waiting for the next review, uncertainty in written follow-up, and competition for limited compensation opportunities.
An “expansion” phase then deepens each choice by generating scenarios—how the boss might respond, what obstacles could arise, what partnerships or resources might be needed, and how unexpected outcomes could be handled. For instance, a written request could trigger a request to meet in person or be escalated to HR; a performance review could lead to agreement or handoff to an HR representative; a direct conversation could be postponed, redirected toward a bonus or non-monetary benefit, or met with a reschedule.
Finally, a “decision” phase ranks the solutions in order of promise and justifies the ordering. The recommended order in the example is performance review first, then written request, then direct conversation. The justification emphasizes alignment with regular compensation discussions, reduced confrontational risk, and the need to prepare clear achievements and talking points. Even the lowest-ranked option remains viable, with the caveat that success depends heavily on rehearsal and confidence-building—especially for someone who avoids conflict.
Cornell Notes
Tree of Thoughts prompting is presented as a structured way for a language model to solve problems by exploring multiple reasoning paths. The workflow runs in four phases: brainstorm several distinct solutions, evaluate each one with criteria like pros/cons and probability of success, expand each option into realistic scenarios and contingencies, then rank the options and justify the final order. In the raise-request example, the model generates three approaches (written request, performance review timing, and direct conversation), scores them, considers what could happen next (including HR escalation or rescheduling), and recommends performance review first. The method matters because it encourages backtracking and more deliberate selection rather than relying on a single attempt at reasoning.
How does Tree of Thoughts differ from a single “chain” of reasoning when solving a problem?
What does the “brainstorm” phase produce, and why does it ask for variety?
What criteria does the “evaluation” phase use to compare options?
Why does the “expansion” phase matter after scoring options?
How does the “decision” phase turn earlier work into a final recommendation?
Review Questions
- If you had to apply Tree of Thoughts to a different workplace request (e.g., flexible hours), what factors would you include in the brainstorm prompt to force genuinely distinct options?
- Which evaluation criteria would most strongly influence your ranking—probability of success, effort, or implementation difficulty—and why?
- What kinds of “unexpected outcomes” should you plan for in the expansion phase to avoid being caught off guard during the real conversation?
Key Points
- 1
Tree of Thoughts prompting improves problem-solving by generating multiple candidate reasoning paths and allowing backtracking instead of committing to a single approach.
- 2
The method runs in four phases: brainstorm distinct solutions, evaluate them with explicit criteria and success estimates, expand each into realistic scenarios and contingencies, then rank the options.
- 3
In the raise-request example, three distinct strategies were produced: written request, performance review timing, and direct conversation.
- 4
Evaluation criteria included pros/cons, effort, implementation difficulty, potential challenges, expected outcomes, plus probability of success and confidence levels.
- 5
Expansion added practical realism by considering HR escalation, rescheduling, and alternative compensation outcomes like bonuses.
- 6
The final ranking favored the performance review route first due to alignment with normal compensation discussions and reduced confrontational risk.
- 7
Direct conversation remained viable but was flagged as the most stressful option, requiring preparation and rehearsal.