Zetta Playbook

Question 1

What are two things AI adoption is determined by?

Zetta

Answer 1

Both trust and risk

Question 2

What are the four different developments in AI? And examples

(Zetta)

Answer 2

AI personalisation: Associated with low risk for incorrect prediction (eg. Book recommendations)

AI Augmentation: Still function without, augment on top of (eg. Improved search suggestion tool)

AI automation: Workflow fully automated by AI (eg. Tractable visual diagnose car crash)

AI Creation: Products not exist at all without AI (eg. Invenia models predict demand and supply of electricity)

Question 3

What is the difference between systems of record to systems of intelligence?

(Zetta)

Answer 3

Old model teams would manually pull data, crunch it, and put in a report

Intelligence: Can access in real time, visualise and analyse on the fly through intuitive products

Question 4

What does building an intelligent system start with?

Zetta

Answer 4

Having the right data

And
-effectively storing you data and complimenting it with data from other sources

Question 5

The more complex a system the more AI is vulnerable to what?

(Zetta)

Answer 5

Hard-coded inefficiencies, more vulnerable to confounding factors

Question 6

Move from cloud computing era to applications that do what?

Answer 6

Help customers to make decisions

Question 7

Shifted from code to what in intelligence era?

Zetta

Answer 7

Unique data and self-learning code

Question 8

The data value chain flow involves?

Zetta

Answer 8

1. Acquiring data
2. Storing the data
3. Annotating the data

Question 9

Where does the strength of data moats come from?

Zetta

Answer 9

Context

Question 10

Investors talk about what two different types of data?

Zetta

Answer 10

Public data Vs proprietary data

Question 11

Zetta outlines 8 elements to a strong data set. What are they?

Answer 11

1. Accessibility
2. Time
3. Cost
4. Uniqueness
5. Dimensionality
6. Breadth
7. Perishability
8. Virtuous loop

Question 12

What two elements determined how time impacts the strength of a data set?

(Zetta)

Answer 12

Can the data be accessed instantly, or does it take a significant amount of time to obtain and process?

How quickly can the data be amassed and used in the model?

Question 13

What two elements of cost need to be considered in regards to the value of a data set?

(Zetta)

Answer 13

How much money is needed to acquire this data?

Does the user of the data need to pay for licensing rights or pay humans to label the data?

Question 14

How does uniqueness determine the strength of a data set? And what are examples?

Answer 14

Is similar data widely available to others who could then build a model and achieve the same result?

Such so-called proprietary data might better be termed “commodity data” — for example: job listings, widely available document types (like NDAs or loan applications), images of human faces.

Question 15

How does dimensionality impact the strength of a data set? And what are they relevant for?

(Zetta)

Answer 15

How many different attributes are described in a data set?

Are many of them relevant to solving the problem?

Question 16

What three questions on breadth impact the strength of a data set

Answer 16

How widely do the values of attributes vary?

Does the data set account for edge cases and rare exceptions?

Can data or learnings be pooled across customers to provide greater breadth of coverage than data from just one customer?

Question 17

What two elements of perishability determine the strength of a data set?

(Zetta)

Answer 17

How broadly applicable over time is this data?

Is a model trained from this data durable over a long time period, or does it need regular updates?

Question 18

How does the virtuous loop play into the strength of a data set? And do what over time?

Answer 18

Outcomes such as

• performance feedback
• predictive accuracy

can be used as inputs to improve the algorithm

And performance cab compound over time

Question 19

What are one the of the most important ways to differentiate? And how can this be truly defensible?

Answer 19

Data sets

What data moats are tied to a particular problem domain, in which unique, fresh, data composes in value as it solves problems for customers

Question 20

What is one fo the biggest costs to overcome in AI for startups?

(Zetta)

Answer 20

Cost of annotation or labelling

Question 21

Interesting example: What is Cloud auto ML google doing?

Answer 21

Enables developers with limited machine learning expertise to train high-quality models specific to their business needs. Build your own custom machine learning model in minutes.

In sight, language, structure, and platform areas

Paid service

Question 22

What is the Snorkel project out of Stanford? And what three programming operations are used?

Answer 22

Snorkel is a system for programmatically building and managing training datasets without manual labeling. In Snorkel, users can develop large training datasets in hours or days rather than hand-labeling them over weeks or months.

Snorkel currently exposes three key programmatic operations:

1. Labeling data, e.g., using heuristic rules or distant supervision techniques
2. Transforming data, e.g., rotating or stretching images to perform data augmentation
3. Slicing data into different critical subsets for monitoring or targeted improvement
   Snorkel then automatically models, cleans, and integrates the resulting training data using novel, theoretically-grounded techniques.