Transform ChatGPT into a Strategic Powerhouse with the Game-Changing ToT Approach

Hey! It’s Ben & Maxi, and today I’m very excited to bring you a cool analysis about the power of good prompting using the ToT approach, plus many applicable nuggets to your business! 🎯

In Today’s Caveminds Journey…

We bring some fire for your weekend!

• 🌳 How the Tree of Thought approach to boost your business’ strategic decisions - full prompts + examples

• 🍏 Step ahead of the competition by tackling these powerful questions after Apple’s WWDC 23

• 📚 Generative AI trainings to share with your team

Alright, let’s get started!

TOPIC OF THE WEEK 💡

You know how you ask ChatGPT for a decisive, expert opinion and it responds with a courteous, but rather vague analysis? 😒

Yes, we all share that frustration (no, it's not just you).

Even though GPT-4 or GPT-3.5 is packed with features and capabilities, when it comes to the crunch, it kinda, sorta hesitates to throw its hat in the ring, especially when you want it to make or illuminate real-life decisions.

Imagine you're teetering on the edge of launching a new product and are torn between three alternatives.

You, very logically, turn to ChatGPT: "Can you evaluate these options based on skills, experiences, cultural fit, and potential contribution to the company?"

And after a fancy and polite analysis it will ALMOST ALWAYS end up like this:

Should I call you Soft-GPT already?

via GIPHY

Well, hold on, because the game's about to change. ✋

You know, it’s all about the prompting, baby – (and some very big brains from Princeton University and Google Deepmind that presented this paper and the concept called “Tree of Thoughts”).

In other words, when asking for an evaluation of the 3 product alternatives, you can get ChatGPT (specifically GPT-4) into replies like this:

And that’s how you can get 10X better responses with good prompting.

Still not buying it? This might help: according to these scientists, researchers and professors paper, the ToT approach left traditional GPT-4 in the dust, solving 74% of tasks compared to a dismal 4% with the standard approach.

Now, who wouldn't like those odds?

We've pushed the envelope further. 📩

Instead of a simple match between GPT 3.5 x ToT and GPT 4, we added GPT 4 ToT to the mix. Let's just say the results were not just interesting, they were mind-blowing. Particularly for business applications.

We're about to spill the beans on how to apply it. Stick around as we dive into three potential use cases, compare results side-by-side, and reveal the secret sauce that'll turn your ChatGPT into a decision-making powerhouse. Ready?

Tree of Thoughts... what's that, you ask? Some new-age meditation technique? 🌳

Well, not quite. ToT it’s an approach we can get the LLM to use to solve problems or answer questions by thinking about different steps it could take, which are called "thoughts". 💭

Each "thought" is like a possible move or decision that the language model could make to get closer to the answer or solution.

Sometimes, the language model might realize that the steps it's taking aren't leading to the right answer.

When this happens, the language model can go back to a previous step and try a different "thought".

Take a look at how it works:

This is called "backtracking". So, the ToT approach is a way for the language model to explore different paths to the solution and change its path if needed.

The secret sauce: how can you apply it? 🥫

First, make sure you propose the topic with this kind of structure:

We’re going to see a full business example soon. For now, know that Dave Hulbert proposes a very simple prompt to start with: