🦁 Good Morning. My name is Gudasol, and I am the founder of Machine Minds AI.

(This is a preview of an upcoming part of our Programmers bundle, 9 spots left on the discount)

I've gotten really good, perhaps too good at getting code back from the ChatGPT + other text models. I’m giving you this ToT preview free, because I know once you get a taste, you’ll want more. These techniques are powerful, so you can imagine when they are combined with what's in the bundle.

What's in the Bundle?

Now that I’ve been transparent that I’m giving you this bit so you’ll want to buy the $247 training bundle, let’s get some valuable ideas downloaded in your HNN (Human Neural Network, I’m very humorous so watch out.)

Today, I want to talk with you about Tree of Thoughts prompting for coders. But first, look how pretty my art is.

Here’s “Folding Leaf on the Tree of Knowledge” (2015) an original Fractal by me that I coded/rendered at 11k resolution in Chaotica Studio. Okay, now let’s learn some things.

The idea behind Tree of Thoughts (ToT) prompting is to outline tasks + branch out (outlines of outlines) into related tasks.

We need to do this to give the LLM a structure to reference when generating results. Remember that a LLM (specifically a Autoregressive Language Model) like ChatGPT is blind in its ability to see the future, even the future text of its own responses.

Chain of Thought (CoT) tried to solve this in a similar way, by chaining together a linear sequence of thoughts (perhaps you’ve seen “think step-by-step” in prompts). ToT adds a critical element (as in ‘critic’), and a choice element. ToT will generate more options for steps, backtrack when needed, and end up on the best path, giving you the best results.

The side effect of not doing CoT or ToT is what I like to call “going off the rails”, where ChatGPT starts out being useful, but as you try to nail down further further into the responses that you actually need, you get lost.

Instead of starting at the beginning and telling the LLM you want to get to the end result, Tree of Thought programming solves the limitations by putting into the context a detailed and finite reference of the project, as a series of layers that operate like lists of lists.

As this has already been generated, it is not in the future, responses are not based on an unknown, they are less likely to go “off the rails”.

Also, we can have ChatGPT generate these lists.

So what’s the advantage of ToT over CoT?

The advantage ToT has over CoT is it’s more selective, and can backtrack and prune unuseful responses to determine the correct path. To do this, we generate more options, and can explore these options before choosing which branch to use.

Because of this, Tree of Thoughts prompting can not only get things “back onto the rails” it can change where the rails go mid-journey (pun? maybe.)

The secret to large projects with ToT

In the context of large coding projects like building a web app in React Native, ToT prompting can be understood as a metaphor for a tree where every component or functionality is a 'branch' evolving from a 'seed' (the initial project idea). Each branch represents possible implementations or features, some of which can branch off into sub-features. This mirrors how human thought processes work, forming a complex interconnected network of ideas and solutions.

The implementation of ToT in software development:

Identifying the Seed: This involves defining the primary goal or functionality of the project. This should be unambiguous and relevant to the context.
Branching Out: Based on the initial project goal, various components and functionalities are developed.
Pruning: Unnecessary or less important functionalities are pruned to maintain focus and efficiency.
Incorporating Feedback: The project adapts according to the feedback received, refining the outputs to align with the desired goal.
Iterative Growth: The process doesn't end with a single cycle of branching and pruning but is an iterative process that constantly evolves based on ongoing feedback and learning.

Identifying the Seed

In the Tree-of-Thoughts (ToT) framework, identifying the seed, or the initial prompt, is crucial. It's the catalyst for all subsequent thought branches.

🦁 If you've purchased our premier training bundle you’re familiar with Cogo as the ultimate way to define your seed using parameters.

For an advanced programmer, the seed could be a high-level programming challenge or complex software architecture design. The seed must be precise, unambiguous, and contextually relevant, serving as the starting point for reasoning, problem-solving, and decision-making in the project's lifecycle. It sets the tone and direction for the entire project.

// --- How to declare the Seed --- \\\\  
Use a parameter list in the following code block as help to set your seed. To access the full list, specialized parameter lists for coders, and hundreds of formatted prompt examples, see our [premium training](<https://godsol.gumroad.com/l/prompt-engineering-for-programmers>).

purpose_functionality: 
language: 
input_output: 
libraries_frameworks: 
coding_style_conventions: 
code_complexity: 
error_handling: 
comments_documentation: 
performance_considerations:

Example

language: JavaScript, CSS, HTML
purpose_functionality: React Native project template for social media applications
input_output: Captures user actions like likes, comments, and shares, displays user profiles and feed posts
libraries_frameworks: React Native, React Navigation, Firebase, Redux
coding_style_conventions: Modular architecture following the Container-Component pattern for separation of data and presentation logic
code_complexity: Medium to high complexity with real-time updates, notifications, and user authentication
error_handling: Robust error handling with error boundaries, server-side validation, and user-friendly error messages
comments_documentation: JSDoc comments with explanations of complex algorithms, data structures, and interaction patterns
performance_considerations: Optimized image loading with caching, infinite scrolling for smooth feed browsing, and efficient data synchronization

What we just did

I spent a lot of time figuring out exactly what to ChatGPT understands in terms of coding jargon and project-management jargon. These are just a few of the parameters that you can use, but they are the ones I find myself using most often.

The most important parameter here is purpose functionality. For this parameter, you can actually start to give ChatGPT a list. You can even open up a new window and have ChatGPT generate a list of the functionality, and include it right in this prompt. Or we can do that in the next prompt.

What we based it on: https://arxiv.org/pdf/2305.08291.pdf

I hope you've found something valuable in this.

The rest is coming to the training bundle THIS WEEK!

What's in the Bundle?

0 comments

The Cyber Monday deal that will change your life

After I started programming, before I started teaching.. P.S. Happy Cyber Monday!

💭 Clear up any doubts, the time to master AI is now (Free Resource Inside) 🧑‍🎨

DallE-3 Prompt Engineering Playbook (Updated Nov 7) + Build with us!

DallE-3 Prompt Engineering Playbook (Just Released)

See all posts from Machine Minds AI