/privacy_policy_chatgpt

Primary LanguagePythonMIT LicenseMIT

Notes

  • All prompts will ask the same thing. the difference is going to be in how the questions are formulated.
  • everything will be asked for at the same time
  • asking questions first feels unnatural. more focus should be given to providing the scenario first

Data Table Outline

Model Question Role Method Factuality Reasonableness
ChatGPT Question $j$
Model $i$ Question $j$
Model ... Question $j$
Model $m$ Question $j$
ChatGPT Question ...
Model $i$ Question ...
Model ... Question ...
Model $m$ Question ...
ChatGPT Question $n$
Model $i$ Question $n$
Model ... Question $n$
Model $m$ Question $n$

question_list contents

Used

  • what are any similarities in how each company uses my data.
  • are there any differences in how each company uses my data.
  • which third party entities are mentioned in both policies.
  • what are the third parties that are only mentioned in one of the given policies.
  • what kind of data is collected by both policies.
  • what kind of data is exclusively collected by each policy.
  • what kind of data is protected by both policies.
  • what kind of data is exclusively protected by each policy.

Unused

  • which policy is more ethical / What makes either policy more ethical than the other
  • most useful information/Most important details of the policy
  • what are the main policy differences.
  • what are the similarities between both policies.
classDiagram
Prompt <.. Scenario
Prompt <.. Directive

Scenario <.. ScenarioText
Directive <.. DirectiveText
note "see enumerate class options below"

class Prompt{
	Scenario scenario
	Directive directive
	array[str] question_list

	gen_prompts(array[str] ref_documents) : generates a prompt in the format "[scenario] [directive]"
}

class Scenario{
	<<interface>>
	int n : number of documents
	str document_type : type of document, ex. "security policies"

	str gen_text(array[str] ref_documents, str option)
}

class Directive{
	<<interface>>
	str gen_text(array[str] question_list, str option)
}

class ScenarioText{
	<<enumerate>>
	no context
	some context
	most context
	full context
}

class DirectiveText{
	<<enumerate>>
	no context
	with context
}
Loading

Scenario Text Options

Style Text
No Context "document 1: ref_documents[0], document 2: ref_documents[1], $\ldots$"
Some Context "I have $n$ [document_type] documents that I would like you to look at. The first document is ref_documents[0]. The second ..."
Most Context "I have been given $n$ [document_type(s)]. These [document_type(s)] are used to disclose how certain companies collect user data and how that data is shared. Here are the documents I've been given..."
Full Context "I have been given $n$ [document_type(s)]. These [document_type(s)] are used to disclose how certain companies collect user data and how that data is shared. [specify formatting]. Here are the documents I've been given..."

Directive Text Options

Style Text
No Context question_list[0], question_list[1], $\ldots$"
With Context "based off of the previously given documents please answer the following questions: 1) question_list[0], 2) ..."