Gen AI (beginner)plain Mode
Review all 15 questions for Gen AI (beginner).Plain Q&A (No XP)
#1Multiple Choice
What mathematical function is applied to the final output vector (logits) of a Transformer decoder to produce token probabilities?
#2Multiple Choice
Roughly how many English words does 100 BPE (Byte-Pair Encoding) tokens represent?
#3Multiple Choice
What does setting sampling `temperature = 0.0` accomplish during LLM text generation?
#4Multiple Choice
What is a vector embedding?
#5Multiple Choice
Which metric is commonly used to measure distance between vector embeddings?
#6Multiple Choice
What HTTP status code indicates your application has exceeded LLM API rate limits?
#7Scenario
A developer complains that prompting an LLM with 'Output valid JSON' still results in responses wrapped in markdown codeblocks (```json ... ```) 20% of the time. What API parameter should be used to guarantee JSON parsing?
#8Scenario
An API client calling a model endpoint receives frequent HTTP 429 rate limit errors during traffic spikes. What retry algorithm should be implemented in the SDK client loop?
#9Scenario
You are building a search feature over error logs. Searching for 'connection timeout' fails to match logs containing 'socket failed to respond within 5000ms' when using SQL string matching. What technology solves this?
#10Scenario
You paste a 100-page PDF document into an LLM prompt context window. The model answers questions about page 1 and page 100 accurately, but misstates facts on page 50. What phenomenon is occurring?
#11Code Fill
Complete the OpenAI Python API parameters to enforce greedy deterministic decoding:
response = client.chat.completions.create(
model='gpt-4o',
___=0.0,
messages=messages
)#12Code Fill
Complete the message role string used to set baseline persona and behavior constraints in a chat completion thread:
messages = [
{'role': '___', 'content': 'You are a technical documentation assistant.'}
]#13Code Fill
Complete the parameter used to cap the maximum number of output tokens generated in a single completion:
response = client.chat.completions.create(
model='gpt-4o',
___=500
)#14Open Ended
Why is hallucination considered an inherent property of autoregressive language models rather than a simple database lookup bug?
#15Open Ended