ANT Thinking in Prompting: New LLM Technique

A new concept called ANT Thinking is getting attention from Artificial Intelligence (AI) and Language Model experts.

It was first found when users looked at the instructions used to communicate with Claude 3.5 Sonnet (Anthropic version of ChatGPT).

These instructions are called system prompts.

Online discussions and leaks on social media, especially on Twitter, have shared interesting info about these prompts.

Online discussions and leaks on social media, especially Twitter, have shared some interesting information about these prompts.

Interestingly, @elder_plinius‘ tweet mentions that Claude “completely bugs” when he tries to repeat or discuss this topic.

So, what is ANT Thinking?

Where ANT Thinking Comes From

ANT Thinking is an approach to improving LLM decision-making.

This method has its roots in psychology, where a well-known concept called Automatic Negative Thoughts (ANTS) was developed by Dr. Daniel Amen.

In human psychology, ANTS refers to the unwanted and unhelpful thoughts that can pop up in our minds, influencing our mood and behavior.

For instance, if someone is feeling anxious, they might start thinking “I’ll never be able to do this” – that’s an example of an ANT.

The idea behind ANT Thinking is to adapt this concept of automatic reflective thoughts to artificial intelligence.

Just as humans can reflect on their own thoughts to improve their mental well-being, AI systems using ANT Thinking can engage in a reflective process to enhance their responses and manage any “mental” artifacts they may have developed.

This helps the AI system to make more informed and optimized decisions.

What Is ANT Thinking?

When we talk about “ANT Thinking” in Large Language Models (LLMs), we’re referring to a specific decision-making process.

“ANT” stands for “Automatic Negative Thoughts”, a concept borrowed from psychology.

In AI, “ANT Thinking” means the system constantly evaluates its potential responses.

This evaluation is like a series of checks.

The AI asks itself:

• Is this response substantial and meaningful?
• Can the user reuse or modify it?
• Does it make sense in the context?

The goal is to generate “artifacts” – self-contained pieces of content like code snippets, documents, or diagrams.

These artifacts should be helpful and relevant to the user.

The Particularity of ANT Thinking in LLMs

How It Works

ANT Thinking operates behind the scenes in LLMs, guiding the AI through a reflective process before generating artifacts.

This involves:

1. Evaluation: The AI assesses whether the content meets the criteria for being an artifact.
2. Decision Making: The AI decides if the content should be a new artifact or an update to an existing one.
3. Creation and Tagging: If deemed appropriate, the AI wraps the content in specific tags (like <antartifact>) with identifiers and types, ensuring consistency and traceability.

Hidden Deliberations

An interesting aspect of ANT Thinking is its hidden nature.

When you interact with a system that uses ANT Thinking, there are many complex mental processes happening in the background that you don’t see.

These processes are like a series of thoughts and decisions that the system makes automatically, without bothering you with the details.

This way, the system can provide you with quick and coherent responses that make sense.

The reason these processes are hidden from view is to make your experience with the system as smooth and easy as possible.

If the system were to constantly stop and ask for your input or clarification, it would be frustrating and annoying.

By handling all these background tasks on its own, the system can provide a more seamless and enjoyable experience.

How ANT Thinking Is Used

Artifact Creation

ANT Thinking is central to creating and managing artifacts in conversations.

It ensures that substantial, reusable content is appropriately encapsulated and identified.

Examples include generating Python scripts, SVG images, and flowcharts.

The AI uses ANT Thinking to decide if a piece of content should be turned into an artifact based on its complexity and potential for reuse.

Updating Artifacts

When users request modifications, ANT Thinking helps the AI determine whether to update an existing artifact or create a new one.

This process maintains the continuity and evolution of content, making it easier for users to follow and iterate on their work.

Why ANT Thinking Is Used

Enhancing Output Quality

The main goal of using ANT Thinking is to make sure the artificial intelligence (AI) system produces high-quality and relevant results.

To achieve this, the AI system goes through a thorough process of examining potential answers to a question or problem.

This careful evaluation helps the AI to generate more thoughtful and meaningful responses, which in turn reduces the chances of producing useless or unimportant information.

User Experience

ANT thinking simplifies interactions.

It ensures that substantial content is presented in a simple way, hiding the intermediate thoughts of the AI.

This approach reduces the cognitive workload for users.

They receive well-structured, easily accessible information, without having to sift through unnecessary detail.

Practical Example of ANT Thinking in Prompt Output

Let’s say a user asks an LLM to create an SVG image.

The AI first evaluates if this task meets the criteria for a good artifact:

User Query:

Can you draw a simple blue circle in SVG?

Assistant Response:

I'll create a simple blue circle for you using SVG.

<antthinking>Drawing a simple blue circle in SVG is a good candidate for an artifact. It's a standalone visual element that can be reused or modified. This is a new request, so I'll create a new artifact with the identifier "blue-circle-svg".</antthinking>

<antartifact identifier="blue-circle-svg" type="image/svg+xml" title="Simple blue circle SVG">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 100 100">
 <circle cx="50" cy="50" r="40" fill="blue" />
</svg>
</antartifact>

Here's a simple SVG image of a blue circle, as requested. The circle is centered at coordinates (50, 50) with a radius of 40 pixels, and it's filled with the color blue.

Key Points from this Article

• AI uses “ANT Thinking” to evaluates responses: When you ask the AI a question, it thinks carefully about how to answer.
• The AI does its “ANT Thinking” behind the scenes: The AI system does a lot of complex work in the background, so you don’t have to worry about the details.
• “ANT Thinking” enhances output quality: By taking the time to think carefully about its responses, the AI system can give you more thoughtful and useful answers.

👉🏻 Diving Deeper into LLM and Prompting

ANT Thinking is a very new and relatively unknown topic of discussion.

There isn’t much information available about it yet.

To write this post I’ve gathered some information from LinkedIn and X. I want to thank @elder_plinius, Philipp Schmid, and Ross F. for their post about the topic.

You can find my own writings on AI and prompting topic on these platforms. Find me there :

• X
• Linkedin