An API key should be treated like:

A private password granting model access

A diffusion model starts generation by:

Adding noise and then learning to remove it

The Five C’s checklist includes Compliance, which means:

Meeting privacy & regulatory requirements

Mixo is primarily positioned as:

A one-page lead-generation site builder

Retrieval-Augmented Generation (RAG) is when an LLM:

Looks up relevant chunks from a knowledge base before answering

PartyRock (AWS) turns a long prompt with {{variables}} into:

A runnable mini-app with user inputs

The smallest text chunk the model actually understands

A system prompt is the message role that:

Sets global rules and tone for the model

ElevenLabs in Session 6 was used to demonstrate:

A voice agent that can speak and act via MCP

A vector store is primarily used to:

Hold embeddings for fast similarity search

Embedding a text chunk produces:

A fixed-length numeric vector representing its meaning

Few-shot prompting means:

Providing a handful of examples inside the prompt

A vector store + RAG solution is cheaper than raw PDF because:

Only the relevant chunks are sent to the LLM

MCP (Model-Context-Protocol) allows an AI agent to:

Call external tools in a standardized way

A Generative Adversarial Network (GAN) is best defined as:

Two neural networks that compete to create and detect fakes

In image prompting, the style/artist element specifies:

The overall visual aesthetic (e.g., watercolor, Van Gogh)

The main purpose of RAG in enterprise chatbots is to:

Reduce hallucinations and token cost simultaneously

In a Transformer, the term attention means:

Weights showing how strongly one token influences another

Ideogram AI was introduced as:

An image-generation tool that defaults to public outputs

The phrase; "context window" refers to:

The maximum number of tokens the model can consider at once

studbud

Sessions 1 - 8

23 termsStudBud Administration28 plays

Imported from external platform

Flashcards23 cardsFPractice quizPHost liveL

Questions 23

Answers

1
An API key should be treated like:
A private password granting model access
A browser cookie you can delete anytime
An open-source license
A public Slack emoji
2
A diffusion model starts generation by:
Adding noise and then learning to remove it
Translating text into vectors first
Drawing random brush strokes until correct
Copying half an existing image
3
The Five C’s checklist includes Compliance, which means:
Counting tokens accurately
Checking CPU usage limits
Meeting privacy & regulatory requirements
Choosing the cheapest model
4
Mixo is primarily positioned as:
A video-avatar generator
A vector-database client
A one-page lead-generation site builder
A voice-cloning toolkit
5
Retrieval-Augmented Generation (RAG) is when an LLM:
Looks up relevant chunks from a knowledge base before answering
Fine-tunes on synthetic images
Switches to agent mode automatically
Generates answers without any outside data
6
PartyRock (AWS) turns a long prompt with {{variables}} into:
A compiled Android binary
A real-time stock dashboard
A GitHub actions workflow
A runnable mini-app with user inputs
7
A token is:
A full English word plus punctuation
A CSS style for UI elements
An encryption key for APIs
The smallest text chunk the model actually understands
8
A system prompt is the message role that:
Sets global rules and tone for the model
Logs inference time
Stores model weights
Supplies user questions
9
ElevenLabs in Session 6 was used to demonstrate:
A voice agent that can speak and act via MCP
Cloud GPU provisioning
AR object detection
Spreadsheet formula generation
10
A conversational model tier is optimized for:
Training new embeddings
Tool use and external actions
Fast, low-cost chat replies
Complex multi-step reasoning
11
A vector store is primarily used to:
Cache CDN assets
Hold embeddings for fast similarity search
Save user form inputs
Render 3-D graphics
12
Uizard / Google Stitch output files most similar to:
Binary model checkpoints
Raw SQL schemas
Docker compose files
Editable Figma mock-ups
13
Embedding a text chunk produces:
A compressed PDF file
An HTML iframe
A fixed-length numeric vector representing its meaning
An anonymized user ID
14
Few-shot prompting means:
Running the model only a few times
Sampling fewer output tokens
Providing a handful of examples inside the prompt
Shortening the context window
15
A vector store + RAG solution is cheaper than raw PDF because:
The API key bypasses billing
JSON compression is disabled
Only the relevant chunks are sent to the LLM
GPU memory is over-allocated
16
Temperature in an LLM API controls:
HTTP request latency
The model’s factual accuracy score
GPU temperature during inference
Output randomness/creativity
17
MCP (Model-Context-Protocol) allows an AI agent to:
Tune temperatures on the fly
Call external tools in a standardized way
Compress models for mobile
Increase context windows to 1 M tokens
18
A Generative Adversarial Network (GAN) is best defined as:
A single network that compresses data
Two neural networks that compete to create and detect fakes
A rule-based image filter
A database of stock photos
19
In image prompting, the style/artist element specifies:
The lighting direction
The camera brand used
The overall visual aesthetic (e.g., watercolor, Van Gogh)
The file resolution
20
The main purpose of RAG in enterprise chatbots is to:
Reduce hallucinations and token cost simultaneously
Increase image generation speed
Allow models to self-fine-tune on the fly
Replace the need for prompt engineering
21
In a Transformer, the term attention means:
A human supervising model output
Extra VRAM for large models
Weights showing how strongly one token influences another
A dropout layer to prevent overfitting
22
Ideogram AI was introduced as:
An image-generation tool that defaults to public outputs
A JavaScript CSS framework
A speech-to-text API
A time-series forecasting library
23
The phrase; "context window" refers to:
A browser tab showing API docs
The maximum number of tokens the model can consider at once
A pop-up with usage tips
The model’s prediction horizon in time-series data

Sessions 1 - 8 Flashcards | StudBud