majeedkazemi / code-aid Goto Github PK

multi-prompt large answer:

• overall-explanation: provide a short explanation/answer to the question
• sub-prompt-1: generate simple example + explain how X is used in it
• sub-prompt-2: generate complicated example with 1 other concept, randomly selected from a list of applications (network, database, file, 3d graphics, image processing, game physics)

• each q+a block to have an optional follow-up text-box
• the reply will be displayed right below it
• a new follow-up text-box will be displayed after the new response

• include all previous answers as part of the prompt

• will report how much time each student spent using this over the semester
• find correlations?

although, how much time is probably highly correlated with how many usages they have (or events in general)

• model-name: code-inquiries

• types: code-explanation, question, code-question, code-breakdown, code-help-fix

• each type is responsible for managing its own structured data.

• something similar to linux man pages
• allow follow-up questions

-> could have a completely different interaction in which the user would first search for the keyword that they're looking for and see it. Then they could use that to ask questions from.

• get the response URL + id

these could simply fill out the tool and press submit

• be temporary (not be stored in the db)

• not require feedback

• have a question:

  • want to know more about how to use a particular function? -> ask question
  • don't understand what a line of code does? -> paste code + ask question + include code that you're asking about

• stuck:

  • don't know why there is an error in a particular line -> paste code + ask question + include problematic code
  • use the help fix functionality

explain code:

• you've been given a starter code, but you don't understand some parts of it -> paste code

https://help.openai.com/en/articles/5955598-is-api-usage-subject-to-any-rate-limits

useful? yes / no [choice | required] + why? [text | optional]

should receive the following:

• what they are trying to do
• error message (optional)
• what the problem is
• their code

codex-router should try to help them with the above input

• create an elasticsearch server
• store indexed data (date, question, answer, code, annotatedCode, and etc. in a way that we could query later)

• most recent responses
• most recent unique users
• histogram
• average
• positive / negative feedback

• check if you could fine tune one model and then just use different prompt messages for different features

codex-router:

• to add comments after each line explaining all the details of each line.
• parsing it in an array

react UI:

• display the code with on-hover interaction

we are not promising they are correct, regardless of their consent we are collecting data (and the instructors will see how the tool is being used)

• relevance: 0, 1, 2

• the rest: 0 - 4

• change the label for revealing: to how detrimental to learning.