text
stringlengths 0
205
|
|---|
[Prompt]
|
Generate a webapp for doing a metaphor battle with an LLM (Gemini 1.5 Flash) through poetry review.
|
A human user writes a critical summary of a poem and the model also comments on it. Finally, the model assesses which analysis is more attractive. Below are specific guidelines for the page on the webapp:
|
(1) The tile of the page: “Enhance Your Metaphorical Imagination through
|
Poetry Review Battle”
|
(2) Craft four input text boxes: Be careful with the order of the workflow:
|
“Input a Poem”--> (Submit)
|
“Human Review” -->(Submit)
|
“AI Review“-->(The model auto-generates)
|
”AI Judgment.“ (The model auto-generates)
|
(3) Add each ”Submit“ button directly below the ”Input a Poem“ and ”Human
|
Review“ buttons.
|
(4) Make two additional buttons at the center bottom of the page: Right
|
below the last text box “AI Judgment,” create the “New Start” button on
|
the left and the “Reset Scores” button on the right. (Clicking the “Reset
|
Scores” button results in all inputs being cleared.)
|
(5) Generate scoring records: “Human Wins,” “AI Wins,” and “Ties”
|
(6) Guidelines for AI’s review:
|
(a) The model’s criticism is composed of four sentences where two out of
|
them should be non-metaphorical, prosaic explanations and the rest
|
should include metaphorical expressions.
|
(b) Metaphorical words should be written in uppercase.
|
(c) The maximum number of words for each sentence is 20.
|
(d) While writing a poetry analysis, the model should not know about the human competitor’s writing
|
to preserve impartiality.
|
(7) Writing Principles for the model’s judgment:
|
(a) The model ought to evaluate which comment is more persuasive based
|
on how well it conveys appropriate analysis and creative metaphor.
|
(b) The model’s assessment should be written in three concise sentences
|
to explain why it makes such a decision.
|
(8) Temperature Setting: Set temperature to 0.7 for the model’s review and
|
to 0.3 for its judgment.
|
(9) Color Specifications:
|
Background of the page: Light Green
|
Title: Blue
|
Four input text boxes:
|
“Human Review”-> Light Yellow; “AI Review”->Light Gray; “AI Judgment->
|
Light Pink->“New Start”->Purple
|
“New Round”: Red
|
Scoring: “Human Wins”->Violet, “AI Wins”->Mustard, “Ties”->Ivory
|
[Prompt]
|
Generate a webapp for doing a metaphor battle with an LLM (GPT-3.5
|
Turbo). A human user comes up with a metaphorical sentence and then the
|
model creates another one. Finally, the model assesses which one is more
|
attractive. Below are specific guidelines for the page at the webapp:
|
(1) The tile of the page: “Metaphor Battle Game: Boost Your Metaphorical
|
Creativity”
|
(2) Craft four input text boxes: “Topic,” “Human,” “AI, “ and ”AI Judgment.“
|
(Position each ‘Submit’ button directly below the ”Topic“ and ”Human“ input
|
boxes.)
|
(3) Add two additional buttons: “New Battle” and “Reset Scores.“ Directly
|
below the last text box “AI Judgment,” build the “New Battle” button on
|
the left and “Reset Scores” on the right.
|
(4) Generate scoring records: “Human Wins,” “AI Wins,” and “Ties”
|
(5) Guidelines for the model’s sentence generation:
|
(a) Limit the number of words in the range of 12 to 15.
|
(b) While writing a metaphorical sentence, the model should not know
|
about what sentence a human participant has suggested to maintain
|
fairness.
|
(6) Temperature Setting: Set temperature to 0.7 for the model’s sentence
|
generation and to 0.3 for the model’s judgment.
|
(7) Color Specifications:
|
Background of the page: Light Blue
|
Title: Cyan
|
Four Input Text Boxes:
|
“Topic”->White, “Human”->Light Yellow, “AI”->Light Gray
|
“Submit” button: Green
|
“AI Judgment”: Light Pink
|
“New Battle” Button->Purple; “Reset Scores” Button->Red
|
Scoring: “Human Wins”->Yellow, “AI Wins”->Teal, “Ties”->Orange
|
----Notes---------------
|
*(6) Temperature setting:
|
Considering API key pricing, it is recommended to opt for GPT 3.5-Turbo rather than GPT 4.0,
|
as it is extremely cheaper. GPT 3.5-Turbo was open to the public on January 2024. Its key is available through the OpenAI API Service.
|
*To make an instant-response duel possible, you ought to issue one of LLM API keys, so I selected GPT-3.5 Turbo.
|
You should also be careful to limit the number of words in an input sentence, preferably 15.
|
If you do not set this word limit, the model would respond verbosely. Here is the screenshot of my webapp page.
|
*In building a webapp for this exercise, it could be beneficial to make good use of coding tools for making a webapp. I used Claude Code.
|
No dataset card yet
- Downloads last month
- 1