- ML Spring
- Posts
- A Guide to Prompt Engineering
A Guide to Prompt Engineering
Leverage the true potential of LLMs! 🚀
Prompt engineering is more of an art than a skill, you realise the power of prompt engineering only once you understand and start using it properly.
Today, we will explore & understand various types of prompting techniques with illustrative examples:
Zero-shot prompting
Few-shot prompting
Chain of thought prompting
Tree of thought prompting
1️⃣ Zero-shot Prompting
Zero-shot prompting refers to the ability of an AI model to generate meaningful responses or complete tasks without any prior training on specific prompts.
Here’s an example👇
2️⃣ Few-shot Prompting
In contrast to zero-shot prompting, few-shot prompting involves training an AI model with only a small amount of data or examples.
This technique allows the model to quickly adapt and generate responses based on limited examples provided by the user.
Here's an example 👇
3️⃣ Chain of thought prompting (CoT)
Chain of thought prompting is a method where user provides prompts in a sequential manner, building upon previous responses.
By following this approach, the AI model can generate more coherent and contextually relevant outputs, mimicking human-like conversation flow.
Here's an example 👇
4️⃣ Tree of thought prompting (ToT)
Similar to chain of thought prompting, tree of thought prompting utilizes branching pathways & encourages exploration over various chain of thoughts.
Users can explore different possibilities or directions within the conversation by structuring their prompts as branches in a tree-like structure.
This technique enables greater flexibility, exploration & backtracking during interactions with the AI model.
Broadly speaking ToT involves two components:
Thought generation
Thought Evaluation
We use ToT for a Game of 24:
It's is a mathematical reasoning challenge, where the goal is to use 4 numbers and basic arithmetic operations (+-*/) to obtain 24.
For example, given input “4 9 10 13”, a solution output could be “(10 - 4) * (13 - 9) = 24”.
(Refer the image below as you read ahead)
To frame Game of 24 into ToT, we decompose the thoughts into 3 steps, where each step is an intermediate equation.
Figure 2(a): Though Generation
Decompose the thoughts into 3 steps, each an intermediate equation.
What happens at each step (tree node):
Extract the “left” numbers.
Prompt the LM (Language Model) to propose possible next steps.
Use the same “propose prompt” for all 3 thought steps.
Note: Only one example with 4 input numbers is provided.
Breadth-first search (BFS) in ToT:
At each step, retain the best b = 5 candidates.
Figure 2(b): Evaluation
Prompt LM to evaluate each thought candidate as “sure/maybe/impossible” with regard to reaching 24.
The goal is to promote correct partial solutions that can be confirmed with few lookahead trials.
Eliminate impossible partial solutions
Retain the rest labeled as “maybe”.
Sample values 3 times for each thought.
Check this out👇
That’s all for today, stay tuned for more amazing stuff coming up on AI Engineering!
Thanks for reading! 🙂