AI Study Cards Flashcards
GANs are made up of what two network types.
What are the two parts called?
How does each part work?
GANs are made up of two neural networks: a generator and a discriminator. The two networks compete with each other, with the generator creating an output based on some input and the discriminator trying to determine if the output is real or fake. The generator then fine-tunes its output based on the discriminator’s feedback, and the cycle continues until it stumps the discriminator.
Transformer Models create …’s based on….. data rather than …. … ….
How does it help the model?
This approach helps the model …. … context and is why it’s used to … or … text.
- Create outputs based on sequential data (like sentences or paragraphs) rather than individual data points.
- This approach helps the model efficiently process context and is why it’s used to generate or translate text.
*Like ChatGPT, (which stands for Chat Generative Pretrained Transformer),
Some types of Generative AI Models
- GANs
- Transformer
- (VAEs) Variational Auto-encoders
- (NeRFs) Neural Radiance Fields
What type of model creates 2D & 3D Images?
NeRFs
Name two neural network models
GANs
VAEs
What does ChatGPT stand for?
Chat Generative Pretrained Transformer
What does LLM stand for?
Large Language Model
What type of model is CodeGen?
What does it do?
It democratizes software engineering by helping…
Is an LLM that turns English prompts into executable code. - Democratizes software engineering by helping users turn simple English prompts into executable code
What is Conversational Ai?
What does it enable?
….’s enabling …. ….’s between … … and … …, via a ….
Conducted … … ….’s native language.
Technologies enabling natural interactions between a human and a computer, via a conversation conducted in the human’s native language.
Some examples of Conversational AI Products
- Chatbots
- Voice Assistants
- Virtual Agents
4 Types of Programming
- Classical
- Automatic
- Interactive
- Conversational AI
Classical programming is:
Traditionally, a programmer factorizes a problem into smaller sub-problems, defines a requirement, then drafts a piece of code, which then is revised until it solves the given problem.
In 1945, this is how the first programmable machine, the ENIAC, was programmed using plugboard wiring.
Today, this is how programs are written using formal languages with higher abstraction such as C, Python, or Java.
The classical, fundamental paradigm of specifying a problem in natural language and iteratively refining a solution in a formal or programming language until the specification is satisfied remains the predominant method of programming today.
Automatic programming is:
Coders … in a …-… …, and a … … …-… code.
Most of today’s popular computer languages are like this; coders write in a higher-level language, and a compiler generates low-level code; this saves time and effort, since we humans don’t have to worry about all the low-level details.
Interactive programming is when the person:
Codes a program (or parts of a program) on-the-fly, while that program is running.
Conversational AI programming:
Combines … .. and … programming.
We call what CodeGen does … … programming.
Combines conversational AI (interactive human-to-machine dialogue) and automatic programming (the system automatically creates the program based on a higher-level language:
we call what CodeGen does conversational AI programming.
The advent of machine learning urges us to rethink the classical paradigm. Instead of a human doing the programming, can a machine learn to program itself, with the human providing high-level guidance? Can human and machine establish an interactive discourse to write a program?
The answer, as our research reveals, is a resounding Yes.
your conversation!)
What is Few Shot Learning?
Few-shot learning is a type of fine-tuning using a small number of task-specific examples in the prompt, enabling the model to perform better on a task. We can already do this with prompt design and the base LLM. We include instructions and sometimes several examples in a prompt. In a sense, prefeeding the prompt with a small dataset that is relevant to the task.
Fine-tuning improves on few-shot learning by training on a much larger set of examples than can fit in the prompt. This extended training can result in better performance on specific tasks. After a model has been fine-tuned, you won’t need to provide as many examples in the prompt. This saves costs and enables faster requests and responses.
Overfitting is?
When a model becomes….
When a model becomes too adapted to the new dataset
How to adjust for overfitting? (Regularization) - 3 Techniques
- Dropout
- Weight Decay
- Layer Normalization
What is a concern of overfitting when fine tuning?
A major concern in fine-tuning is when a model is trained too closely on a small dataset. It might perform exceptionally well on that dataset but poorly on unseen data.
What is Catastrophic Forgetting?
Incorrect fine-tuning might cause the model to “forget” some of its previous general knowledge, making it less effective outside the specialized domain.
4 Types of Dataset Bias:
- Selection
- Sampling
- Label
- Historical
What is Selection Bias?
The … selected for …-… does … … the … … of the … space.
The data selected for fine-tuning does not represent the full diversity of the problem space.
What is Sampling Bias?
When .. is … in a way that some… .. .. … …. Are less … to be … than …
The data is collected in a way that some members of the intended population are less likely to be included than others.
What is Label Bias?
The … or labels provided in the fine-tuning dataset are … by … …s or ….
The annotations or labels provided in the fine-tuning dataset are influenced by subjective opinions or stereotypes.
What is Historical Bias?
The data … … or … …s that are … … or …
The data reflects historical or societal inequities that are inherently unfair or problematic.
What happens if you use the wrong Hyper-parameter Selection settings while fine tuning?
The wrong hyperparameter settings used while fine-tuning can hinder the model’s performance or even make it untrainable.
Dataset Splitting is when you separate, or partition your data into what 3 sets?
1.training
2. validation
3. test sets.
The model trains on the training set, hyperparameters are tuned using the validation set, and performance is evaluated on the test set.
What are the first 3 steps to prepare your dataset for fine tuning?
Data Collection
Data Cleaning
Dataset Splitting
What are the 3 parts/or steps of Data-splitting?
- Training
- Validation
- Testing
Model is trained on what type of set?
Training set
Hyperparameters are trained on what type of set?
Validation Set
Some examples of LLM architecture are:
GPT-3.5
Bert
RoBERTa
Things to consider when choosing a model:
Whether if fits your specific task
Input & Output size of the model
Dataset size
Technical infrastructure is suitable for the computing power required for fine tuning.
What does architecture selection allow you to do in fine tuning?
Adjust certain components depending on the task.
What types of architectural components can you adjust in fine tuning?
Final Layer for classification tasks
4 Techniques for Monitoring & Evaluating your model are?
Track Loss & Metrics
Early Stopping
Evaluation Metrics
How do you adjust for performance after fine tuning a model?
Calibrate &
Create a feedback loop
Predictive AI Analyzes?
Historical data to predict future possible outcomes.
Generative AI Focuses its model to be able to generate: (4 types of things).
Generate images, texts, video, and even software code based on user input, demonstrating its potential for creative applications.
Difference between Generative and Predictive AI
At their foundation, both generative AI and predictive AI use machine learning. However, generative AI turns machine learning inputs into content, whereas predictive AI uses machine learning to determine the future and boost positive outcomes by using data to better understand market trends.
NLU is the acronym for…
Natural Language understanding
When planning bot conversations which topics are important to focus on?
Context
Personality
Conversation Design
Which terms are important to know before you build your bot?
Variables
Dialogs
Dialog Intents
Entities
Which License do you need for bots?
Service Cloud
To get started with bots what should you set up first?
Service Cloud
Salesforce Sites
Chat Guided Setup Flow
What is Artificial intelligence (AI)?
A branch of computer science in which computer systems use data to draw inferences, perform tasks, and solve problems with human-like reasoning.
Bias is:
… and …. … in a computer system that create … … in … … from the … function of the system, due to … … in the … … process.
Systematic and repeatable errors in a computer system that create unfair outcomes, in ways different from the intended function of the system, due to inaccurate assumptions in the machine learning process.
Corpus is?
A large collection of textual datasets used to train an LLM.
Domain adaptation is the process through which?
The process through which organization-specific knowledge is added into the prompt and the foundation model
Fine-tuning
Is the process of … a …-… … for a … … by training it on a …, …-… ….
The process of adapting a pre-trained language model for a specific task by training it on a smaller, task-specific dataset.
Generative AI gateway in Salesforce is sometimes referred to as:
Alternate terms:
Einstein gateway, the gateway
Generative Pre-Trained Transformer (GPT) is a family of …. … That’s trained on a … … of text … so that they can … …-… …
A family of language models that’s trained on a large body of text data so that they can generate human-like text.
Grounding is the process through which …-… … and … … is added to the prompt to give … … the … it needs to… more ….
The process through which domain-specific knowledge and customer information is added to the prompt to give the model the context it needs to respond more accurately.
Hallucination is a type of … where the model generates … … text that is … … or makes … to … …, given the …
A type of output where the model generates semantically correct text that is factually incorrect or makes little to no sense, given the context.
HITL stands for?
Human In the Loop
HITL requires?
Human Interaction
A model that requires human interaction
What does Hyperparameter do?
And where do they sit?
A parameter used to control the training process. Hyperparameters sit outside the generated model.
Inference is the process of?
The process of requesting a model to generate content.
Inference pipelines are a sequence of …. … …. ?
They are … together to … … … …
This includes these 4 things:
Resolving … ….
Passing .. … an …
Moderating … …
And sending … … to the …
A sequence of reusable generative steps stitched together to accomplish a generation task. This includes resolving prompt instructions, passing it to an LLM, moderating the results, and sending results back to the user.
What is Intent?
An end user’s goal for interacting with an AI assistant.
What does a Large language model consist of? (LLM)
A language model consists of a neural network with many parameters trained on large quantities of text.
Machine learning is a ?
It specializes in ?
These are designed to ?
A subfield of AI specializing in computer systems that are designed to learn, adapt, and improve based on feedback and inferences from data, rather than explicit instruction.
Model cards are?
What 5 things does it include?
Documents detailing about the model’s performance.
It includes inputs, outputs, training method, conditions under which the model works best, and ethical considerations in use.