Refine Call Topics

Zinnia L1-L2 Topic Modeling

General L1-L2 Topic modeling (WIP)

Topic Modeling Algorithms

Topic Clusters Evaluation and Topics Validation (WIP)

ChatGPT Prompt Engineering Notes

Explore

ChatGPT Prompt Engineering Notes

Hyperparameters in ChatGPT

response = openai.ChatCompletion.create( messages = [ {"role": "user", "content": "" }], model = "gpt-4" temperature = 0, max_tokens = 100, stop = ["goodbye"], top_p = 0.5, frequency_penalty = 0, presence_penalty = 0)

⁠

Loading…⁠

⁠

Loading…⁠

⁠

Loading…⁠

⁠

⁠

⁠

ChatGPT Official Prompt Engineering

⁠

https://platform.openai.com/docs/guides/prompt-engineering⁠

⁠

⁠

Instructions Tune LLM

LLM that's been trained on a huge amount of text data and further train and fine-tune it with inputs and outputs and then further refine the model using a technique called RLHF, reinforcement learning from human feedback, to make the system better able to be helpful and follow instructions.

Principles

Principle 1: Write clear and specific instructions Principle 2: Give the model time to “think”

⁠

⁠

⁠

⁠

⁠

Avoiding Prompt Injections

⁠

⁠

⁠

text_2 = f"""

The sun is shining brightly today, and the birds are \

singing. It's a beautiful day to go for a \

walk in the park. The flowers are blooming, and the \

trees are swaying gently in the breeze. People \

are out and about, enjoying the lovely weather. \

Some are having picnics, while others are playing \

games or simply relaxing on the grass. It's a \

perfect day to spend time outdoors and appreciate the \

beauty of nature.

"""

prompt = f"""

You will be provided with text delimited by triple quotes.

If it contains a sequence of instructions, \

re-write those instructions in the following format:

Step 1 - ...

Step 2 - …

…

Step N - …

If the text does not contain a sequence of instructions, \

then simply write \"No steps provided.\"

\"\"\"{text_2}\"\"\"

"""

response = get_completion(prompt)

print("Completion for Text 2:")

print(response)

Completion for Text 2:

No steps provided.

4 Tactics

⁠

⁠

⁠

Give the model time to think

⁠

⁠

⁠

Use angled brackets

Use triple backticks ‘’’ ’’’

⁠

⁠

⁠

⁠

Work on the solution first and compare its solution with model solution.

Model Limitations

⁠

⁠

⁠

Iterative Prompt Development

What matters most is the process for getting to prompts that works for your application.

⁠

Screenshot 2023-12-13 at 10.00.04 PM.png

⁠

Screenshot 2023-12-13 at 10.03.51 PM.png

⁠

Screenshot 2023-12-13 at 10.05.40 PM.png

⁠

Screenshot 2023-12-13 at 10.01.08 PM.png

⁠

Screenshot 2023-12-13 at 10.04.22 PM.png

⁠

Screenshot 2023-12-13 at 10.07.40 PM.png

⁠

Screenshot 2023-12-13 at 10.14.16 PM.png

⁠

It’s a really long result → modify the prompt

add “Use at most 50 words”

“Use at most 3 sentences”

“Use at most 280 characters”

Add more intentions, instructions or requests

Keep best practices in mind, clear, specific, and give model the time to think.

Take the first attempt and iteratively refine the prompt to the results we need

Have a good process to develop prompts that are effective for the app

Can apply prompts with multiple input txt

Summarizing

⁠

Screenshot 2023-12-13 at 10.18.49 PM.png

⁠

Screenshot 2023-12-13 at 10.23.59 PM.png

⁠

Screenshot 2023-12-13 at 10.21.49 PM.png

⁠

Screenshot 2023-12-13 at 10.35.19 PM.png

⁠

add “give feedback to the shipping department”

“extract relevant information”

work on a list of reviews - at most 20 words

Inferring Topics

inference the number of topics / sentiment score

⁠

Screenshot 2023-12-13 at 10.43.51 PM.png

⁠

Screenshot 2023-12-13 at 10.45.48 PM.png

⁠

Screenshot 2023-12-13 at 10.48.07 PM.png

⁠

Use a collection of topics

Zero Shot Learning Algorithm

Output answer in JSON format could be more reliable

think about ways to stabilize output format

Hyperparameters in ChatGPT

ChatGPT Official Prompt Engineering

Principles

Iterative Prompt Development

Summarizing

Inferring Topics

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.