Large Model

Foundations and concepts

Prompt

test translation Fanyi: prompt words

Instructions to a machine, similar to a programming language.

Token

pending

LLM

test translation Fanyi: Big Model

Essence: A set of Algorithm, similar to a Function.

Common Model

• OpenAI: GPT-3.5/GPT-4/GPT-4V/DALL.EWhisper
• Meta: LLama2 (Open-Source)

GPT Model

The Illusion ofLLM

Because the amount of knowledge the Model faces during the Training process is very large, it cannot Perfecto remember all the info it has seen. An obvious problem is that theModel and if you do not clear about its own Knowledge boundaries.

This means thatwhen answering some obscure Topic hour, the Model may fabricate answers that sound credible but are and if you do incorrectly. This kind of fabricated answer is called "illusion."

For example, in the following example, when we ask:

❝

Tell me about Boy's AeroGlide Ultra Slim smart toothbrush

Among them, the company name exists, but the product name is fictitious. In this case, the Model will still give a fairly realistic fictional product Description.

There are 2 Policy that can be used to reduce the occurrence of this illusion:

• Policy 1: Require the Model to find correlation references and if you do answer questions based on the provided text
• Policy 2: Trace answers back to source files

Temperature

Fanyi: Temperature

This is a common Parameter of a Model. Available values are: 0~1.

• When Temperature is 0 hour: it means the answer is more accurate and fixed, and is suitable for Expectation to get Same output Result every time

• When the Temperature is 0.7 hour: it means the answer is more Stochastic, more creative, and suitable for Expectation to get different output results every time

For example, when answering the hour question "My favorite food is...", the possibility of different foods appearing is different.

When Temperature is 0 hour, the Model always selects the most likely one, namely pizza.

When the Temperature is 0.3 hour, the Model is likely to choose other foods with lower possibility.

When the Temperature is 0.7 hour, the Model is likely to choose other foods with lower possibility.

Purpose of LLM

have common sense

identify intention

classification

Step by step: State machine

Parameter of Supplementary Function (awesome)

Common Prompt Sentences

sum up

Summarize

delimited

Summarize the text delimited by triple quotes.

"""insert text here"""

as follows

If applicable

You will be provided with meeting notes, and your task is to summarize the meeting as follows:
    
    -Overall summary of discussion
    -Action items (what needs to be done and who is doing it)
    -If applicable, a list of topics that need to be discussed more fully in the next meeting.

background

You will be provided

You will be provided with xxx (delimited with XML tags) about xxx topic. 
First xxx. 
Then xxx and xxx.

substep

Use the following step-by-step instructions to respond to user inputs.

Step 1 - The user will provide you with text in triple quotes. Summarize this text in one sentence with a prefix that says "Summary: ".

Step 2 - Translate the summary from Step 1 into Spanish, with a prefix that says "Translation: ".

Consistency

Answer in a consistent style.

Q: Teach me about patience.
A: The river that carves the deepest valley flows from a modest spring; the grandest symphony originates from a single note; the most intricate tapestry begins with a solitary thread.
Q: Teach me about the ocean.

limiting length

Divided into 3 minute points

Summarize the text delimited by triple quotes in 3 bullet points.
Summarize the text delimited by triple quotes in 2 paragraphs.
Summarize the text delimited by triple quotes in about 50 words. // 3 个要点

limited field

very good

Draft a company memo to be distributed to all employees. The memo should cover the following specific points without deviating from the topics mentioned and not writing any fact which is not present here:
xxxx

classification

You will be provided with a tweet, and your task is to classify its sentiment as positive, neutral, or negative.

classification Intent

You will be provided with customer service queries. Classify each query into a primary category and a secondary category. Provide your output in json format with the keys: primary and secondary.

Primary categories: Billing, Technical Support, Account Management, or General Inquiry.

Billing secondary categories:
- Unsubscribe or upgrade
- Add a payment method
- Explanation for charge
- Dispute a charge

Technical Support secondary categories:
- Troubleshooting
- Device compatibility
- Software updates

Account Management secondary categories:
- Password reset
- Update personal information
- Close account
- Account security

General Inquiry secondary categories:
- Product information
- Pricing
- Feedback
- Speak to a human

character and task

your task is to xxx it in a concise way.

concise

explain it in a concise way.

Objective user

To a second grader

Summarize content you are provided with for a second-grade student.

format

output in dot list format

Provide your answer in bullet point form. 

output

• Easy to use
• Provides good value for the price
• High quality and durability
• Difficult to transport
• Difficult to store

ordered list

a numbered list

create a numbered list of turn-by-turn directions from it.

supported formats

format type	example
numbered list	1. Open your browser 2. Input the URL 3. browsing content
List of dots	- Apple Inc-Banana-Laranja
form
Code Block	pythonprint("Hello, world! ")
headings and subheadings	##Main title ###Subtitle
JSON format
Mind Graph Format
markdown format