A New Method for Prompting GPT4

How to improve the quality of 'responses' provided by GPT4? As great as it is, and even as it continues to be refined, researchers continue to study the possibility of improving the 'prompt precision' (ie Brian Roemmele - known for his use of 'hypnagogic' SuperPrompts). But maybe the entire 'prompt structure' could also be improved, which is what researchers proposed with the now-super-popular Tree of Thoughts (ToT):

https://arxiv.org/abs/2305.10601

“Our experiments show that ToT significantly enhances language models’ problem-solving abilities on three novel tasks requiring non-trivial planning or search: Game of 24, Creative Writing, and Mini Crosswords. For instance, in Game of 24, while GPT-4 with chain-of-thought prompting only solved 4% of tasks, our method achieved a success rate of 74%”.

How it works:

With ToT, AI generates a multitude of thought branches for each problem. It evaluates the feasibility of each branch, chooses the best ones, and navigates this 'thought tree' much like the way the human mind explores different paths when solving a problem.

How to use it in basic GPT4 prompting:

Phase 1: Brainstorm Phase - Generate a diverse range of solutions:

PROMPT: "I have a problem related to [problem area]. Could you brainstorm 3 distinct solutions? Please consider a variety of factors such as [optimal factors]."

Phase 2: Evaluation Phase - Objectively assess the feasibility and potential success of each option:

PROMPT: "For each of the 3 proposed solutions, evaluate their potential. Consider their pros and cons, initial effort needed, implementation difficulty, potential challenges, and expected outcomes. Assign a probability of success and a confidence level to each option based on these factors."

Phase 3: Expansion Phase - Delve deeper into each idea, refining and imagining implementations:

PROMPT: "For each solution, deepen the thought process. generate potential scenarios, strategies for implementation, any necessary partnerships or resources, and how potential obstacles might be overcome. Also, consider any potential unexpected outcomes and how they might be handled."

Phase 4: Decision Phase - Rank the solutions in order of promise:

PROMPT: "Based on the evaluations and scenarios, rank the solutions in order of promise. Provide a justification for each ranking and offer any final thoughts or considerations for each solution."

Phase 5: Summary

PROMPT: "Remove the 2 lowest rated ideas and keep and rewrite a compressed summary of ONLY the highest rated idea including the probability score."

Then you can loop:

--loop Phase 1: Brainstorm Phase (loop 2)

PROMPT: "Problem= [repeat the problem here]

Winning idea= [list the summary of the winning idea in Phase 5 above]

Could you brainstorm 2 new creative distinct solutions to the [Problem] and include the following as a 3rd idea: [Winning Idea] so we now have 3 ideas. Please consider the 3 most important factors that will have an impact on the outcome."

--loop Phase 2: Evaluation Phase (loop 2)

--loop Phase 3: Expansion Phase (loop 2)

--loop Phase 4: Decision Phase (loop 2)

--loop Phase 5: Summary (loop 2)

--loop-more Phase 1: Brainstorm Phase (loop 3)

--loop-more etc...

Final Phase: Deepen the Thought Process

PROMPT: "For the winning solution, deepen the thought process. Generate potential scenarios, strategies for implementation, any necessary partnerships or resources, and how potential obstacles might be overcome. Also, consider any potential unexpected outcomes and how they might be handled."

‍