A Roadmap Discussion

Speakers

Sheng Zhao Loc Nguyen

Augmented Experiences Customer Success
Product Owner, Vice President Product Management, Vice President
Agenda
Topic

LLM Powered Application

• What is ChatGPT and LLM
• Breaking Down Barriers with LLM in BI
• Prompt Engineering and Limitations
• MicroStrategy LLM Integration
• Deployment: Data Privacy and Costs

MicroStrategy LLM Solutions and Roadmap

• Enhancing BI Workstreams with AI
• Demos
• Conclusions
Q&A
What is ChatGPT?

Image Source: Unsplash

LLM and ChatGPT

• GPT stands for generative pre-trained

transformer which is a type of large language
model (LLM) neural network.

• A Large Language Model is a deep-learning

algorithm trained on enormous amounts of text
data.

• LLM has billions of parameters

Course (MIT Introduction to Deep Learning | 6.S191) video link:

https://www.youtube.com/watch?v=7sB052Pz0sQ&list=PLtBw6njQRU-
rwp5__7C0oIVt26ZgjG9NI
LLM and ChatGPT: Current Competitive Landscape

• Accelerating Innovation

• Reducing Costs

And many more…

Breaking Down Barriers with LLM in BI

- ChatGPT when asked to make a quote about itself

Why it matters:
• The Power of Questioning: At the core of BI is the ability to ask powerful questions of
your data to gain insight and make informed decisions.
• Frictionless Data Interaction: AI language models helps to reduce the friction between
users and data.
Prompt Engineering: Anatomy of Design
What is Prompt Engineering? Prompt Technique Example for Content Generation

Your task is to help a marketing team create a description for a retail website of a
Task product based on a technical fact sheet.

Write a product description based on the information provided in the technical

Constraint specifications delimited by triple backticks.

The description is intended for furniture retailers, so should be technical in nature and
Context focus on the materials the product is constructed from.

After the description, include a table that gives the product's dimensions. The table should have two
Response columns. In the first column include the name of the dimension. In the second column include the
Format measurements in inches only.

Also append a paragraph that represents the same table data. Example:
Examples "Dimensions, Measurements
Width, 1
Height, 2"
https://learn.deeplearning.ai/chatgpt-prompt-eng/
Prompt Engineering
AI Generated Result

gpt-3.5-turbo
Prompt Engineering: Limitations
What is a Large Language Model not good at?

Response from GPT 3.5

Here are the profit figures for each subcategory:

• Audio Equipment: $51,349
• Cameras: $205,012
• Electronics - Miscellaneous: $139,653
• TV's: $41,562
• Video Equipment: $212,789

Response from GPT 4

I have calculated the profit for each subcategory below:

1. Audio Equipment: $56,719
2. Cameras: $186,826
3. Electronics - Miscellaneous: $123,370
4. TV's: $44,557
5. Video Equipment: $187,530
 Prompt Engineering: Limitations – Update May 3rd
What is a Large Language Model not good at?

GPT 3.5 GPT 4

MicroStrategy LLM Integration
Prompt Engineering process in MicroStrategy and how to address LLM limitations with our Semantic Graph model

MicroStrategy MicroStrategy AI MicroStrategy

User Category Electronics, Assistant Category Sales Semantic
Movies and Books Electronics $3M Graph
have the highest Sales
Movies $2M
in year 2023
Books $1M

Which three
categories have the Action: create a MSTR viz
Engineered Attributes: Category
top sales in year Prompt Metric: Sales
2023? Filter: Year = 2023
Sort: Sales

LLM
Deployment: Data Privacy
Data Privacy on OpenAI and Azure OpenAI

• On April 25, 2023 OpenAI provided an option • No prompts or completions are stored
to turn off chat history in ChatGPT. in the model during these operations
• ChatGPT Business subscription will be • Prompts and completions are not
released for professionals to provide greater used to train, retrain or improve the
control over enterprise data and end users. models.

Source: https://openai.com/policies/api-data-usage-policies

Source: https://learn.microsoft.com/en-us/legal/cognitive-services/openai/data-privacy
Deployment: Costs
How much does it cost to adopt an AI-powered application?

Model Provider Token Prompt Completion C Average Tokens Yearly No. of Total Annual
Limit Cost per 1k ost per 1k Used per use Usage User Cost
tokens tokens case
gpt-3.5- OpenAI / 4k $0.002 2k 100 1k $400
turbo Azure
OpenAI 1000 $4,000
gpt-4 OpenAI / A 8k $0.03 $0.06 3k Prompt + 1k 100 1k $15,000
zure Open Completion
AI 1000 $150,000
gpt-4-32k OpenAI / A 32k $0.06 $0.12 12k Prompt + 100 1k $120,000
zure Open 4k Completion
AI 1000 $1,200,000
•1 token ~= 4 chars in English •1-2 sentence ~= 30 tokens
•1 token ~= ¾ words •1 paragraph ~= 100 tokens
•100 tokens ~= 75 words •1,500 words ~= 2048 tokens

Source: https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them
https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/
MicroStrategy + AI = A Roadmap Discussion
Enhancing BI Workstream with LLMs
Elevating the user experience and analytics

BUSINESS USERS ANALYSTS ADMIN/ARCHITECT

Deeper Insights Advance data exploration Efficient platform management and

and ease of authoring ease of Semantic Graph design
Enhancing BI Workstream for Business Users
Deeper Insights

Automated Insights: With MicroStrategy Insights,

it’s now easy to track data points, runs advanced
models to spot outliers and trends to deliver useful
insights
Enhanced Insight (NLP/NLQ) Powered by AI
Enhancing BI Workstream for Analysts
Advance data exploration and ease of authoring

Typical Self-Service / Data Exploration Process in MicroStrategy:

Data Connection Query & Refine Model Dataset Visualization

Data Creation Powered by AI
FFSQL Generation Powered by AI
Data Wrangling Powered by AI

Remove any rows where the State values don't match one of the 50 states in the
United States
Lightweight Authoring Powered by AI
Lightweight Authoring Powered by AI
Enhancing BI Workstream for Admins/Architects
Efficient platform management and ease of Semantic Graph design
Admin (Workflow Generator) Powered by AI
Architect (Semantic Graph) Powered by AI
Conclusion
• The integration of LLMs with the MicroStrategy BI
platform represents a major step forward, allowing us
to unlock new levels of innovation that were once
unimaginable.

• Harnessing AI allows us to be more productive,

gain deeper insights and make better decisions.

• Let’s embrace this transformative power of AI

together to create a brighter, more data-driven
future!
 Experience the ONE platform.
https://www.microstrategy.com/en