Q1. Multiple choice · 1 point
Final reviewed version
Which statement correctly describes how human intelligence and artificial intelligence work?
Select a multiple-choice option, or enter a short answer and check your answer.
Real video file upload case
Create a review quiz from a lecture video file
We uploaded an AI lecture video and compared the generated questions with its transcript.
Audio was extracted from a user-owned lecture file and transcribed with Whisper-1. We checked the resulting quiz draft against the transcript and replaced one question to prepare the final review quiz.
Published · Updated
QuizGen case
Create a review quiz from a lecture video file
3 multiple choice + 2 short answer1 / 5
Question preview
Which statement correctly describes how human intelligence and artificial intelligence work?
1AI imitates intelligent behavior through input-output relationships, but does not implement an established explanation of how human intelligence works.
2AI first established how human intelligence works and then implemented those principles.
Source → quiz
Source material and generation settings
Review the source material and generation conditions used in this case.
- Source and input
- LLM의 원리 및 활용 - 마르시스 강연 · Original video file uploaded directly
- Transcription
- Whisper-1 API
- Final quiz
- 5 questions: 3 multiple choice + 2 short answer · 1 point each, 5 total
- Review and editing
- Manual transcript comparison · 1 of 5 questions replaced
- Permission
- Rights holder authorized use and web publication
From source to result
From a video file to the final review quiz
Video file upload → audio extraction → Whisper-1 transcription → quiz draft from the transcript → human comparison against the transcript → one question replaced → final review quiz. The questions were based on the audio transcript.
- 01
Upload the video file
The user uploaded an actual lecture video file to QuizGen.
- 02
Extract audio
Audio was extracted from the video. This case did not analyze video frames or slide images.
- 03
Transcribe with Whisper-1
The Whisper-1 API converted the lecture audio into text.
- 04
Generate a draft
Five draft questions were generated from the transcript.
- 05
Compare with the source
A person checked the answer evidence for all five generated questions in the transcript.
- 06
Replace one question
A segmentation question was replaced with a next-word prediction question to better represent the lecture.
- 07
Prepare the final quiz
Explanations were refined, and two brief concept-response questions were presented as short-answer questions.
Generated output
The five questions after review and editing
This is the final reviewed quiz. Questions, choices, and answers for Q1–4 were retained from the selected output, with refined explanations. A person replaced Q5. The export labels its two non-multiple-choice questions as “에세이” (essay); the final quiz presents these brief concept responses as short-answer questions.
Select a multiple-choice option, or enter a short answer and check your answer.
Q2. Multiple choice · 1 point
Final reviewed version
Which statement correctly describes how a conversational model uses earlier conversation in later answers and what that means?
Select a multiple-choice option, or enter a short answer and check your answer.
Q3. Multiple choice · 1 point
Final reviewed version
Which statement correctly describes the six prompt components and their roles?
Select a multiple-choice option, or enter a short answer and check your answer.
Q4. Short answer · 1 point
Final reviewed version
An instruction clearly states what the model should do, while context provides background information. Name the missing component in this prompt. “Greek yogurt is high in fiber and protein. The audience is children. Use a warm, friendly tone.”
Select a multiple-choice option, or enter a short answer and check your answer.
Q5. Short answer · 1 point
Final reviewed version
According to the lecture, what basic operation does an LLM repeat when generating a sentence?
Select a multiple-choice option, or enter a short answer and check your answer.
Lecture video: LLM principles and applications
Watch the public version of the lecture used in this case study without leaving this page.
One question replaced: segmentation to LLMs
We found support in the transcript for all five generated questions. However, the final segmentation question was judged less representative of the lecture as a whole, so a person replaced it with a question about the basic generation principle of LLMs. This was an editorial choice to improve coverage, not a correction of a wrong answer.
Before · Q5 in the selected output
How does segmentation represent image regions in its output?
Answer: Assigning a numeric category label to each pixel
Key evidence in the transcript · original Korean excerpts
영역이 잘라지는게 아니라 그 픽셀에 특정한 숫자를 그냥 assign 하는거죠.
Review result: The original answer is also supported by the transcript. The replacement reflects the relative importance of the review topic.
After · Q5 in the final quiz
According to the lecture, what basic operation does an LLM repeat when generating a sentence?
Answer: Predicting the next word from context
Why we chose this lecture
We chose a long lecture used in an educational setting to review both its conceptual explanations and examples of practical use.
- The lecture discusses human intelligence and AI, conversation context, prompt components, and how LLMs generate sentences.
- The lecture is also available on YouTube. For this case, we directly uploaded the original video file with permission from its rights holder. We did not use the YouTube URL input feature.
- We compared a draft based on the audio transcript with the final quiz edited by a person.
The Whisper transcript was not a polished lecture script
The live recording included recording and display-connection conversations, audience questions, and some misrecognized technical terms and names.
- Some ChatGPT, LLM, and prompt-related terms were transcribed differently. We did not turn guessed spellings into official names or corrected direct quotations.
- The generated questions focused on the lecture concepts rather than incidental conversation. We checked the evidence for each answer in the transcript.
- The supplied transcript has no timestamps, so none are shown. Evidence excerpts retain the original Korean wording, including transcription errors; the quiz below is an English translation of the final Korean quiz.
A brief transcript excerpt · original Korean wording retained
예시하고 펜소나를 주면 퀄리티가 확 좋아지고 형식 어조는 완성도를 높이는 데 쓴다.
What this case showed
This single case illustrates generating a draft, comparing it with the transcript, and editing it into review questions.
- We could create draft questions focused on key concepts in a long lecture transcript.
- Despite incidental conversation and some transcription errors, the transcript contained evidence for the answers to all five generated questions.
- Human review and editing produced a final quiz for reviewing the lecture.
- Quiz generation in this case was based on the audio transcript. It did not analyze visual slide information absent from the spoken audio.
Try the same workflow
Create a review-question draft from your lecture file
Start with your lecture video, check the evidence against the original material, and edit the questions for your review goals. The generator guide explains video file input.