An LLM? What's That?
What does LLM stand for?
Large Language Model
What does an LLM learn from during training?
Text
When an LLM generates an answer, what does it predict?
What comes next
What is a cutoff date?
The point in time after which the LLM was not trained.
True or false: Luckily, an LLM is able to automatically check whether what it is saying is true or not.
False.
An LLM is a computer program trained to read and write what?
Human language
Name one type of source that can be part of an LLM's training data.
Web pages, books, or articles
An LLM doesn't generate an entire response all at once. What does it do?
It generates it one small chunk at a time.
Why might an LLM not know about a recent event?
Because the even happened after it was trained.
Without add-on sign-in features, does an LLM remember every previous conversation you had with it?
No.
An LLM is trained on an enormous amount of what?
Text
An LLM doesn't memorize a list of facts. It learns ___.
It learns patterns.
What is a token?
A small chunk of text that an LLM processes.
If something happened after its cutoff date, how would the LLM come up with an answer?
It would guess or make it up.
By default, does an LLM have live access to the internet?
No.
An LLM is a ___ system.
A mathematical system
The model calculates: What is the most likely ___ of this text, based on my training?
Continuation
A token can be a word, part of a word, or what other type of thing?
A punctuation mark
You should always verify important facts the LLM gives you like names, dates, statistics and ___ .
Quotes
An LLM has no truth-checking, no memory, no live internet, and no ______.
Intent.
What makes an LLM different from a traditional search engine?
An LLM does not simply look up answers; it uses patterns learned during training to generate responses.
Why can an LLM write fluently about a topic but still give an answer that is confidently wrong?
Because it learns patterns in language rather than automatically checking whether the information is true.
According to the lesson, how many tokens are in the word “Learning”?
Two: “Learn” + “ing”
What does the "G" in ChatGPT stand for?
Generative
An LLM can sound very confident while giving incorrect information. What is this kind of incorrect but plausible output called?
A hallucination