AI-powered profile summary helper

[steveasher]

Integrate with ChatGPT to support rewriting a profile description. This is very similar to the "Rewrite with AI" feature in LinkedIn (Premium).

[RobotGR]

@steveasher @frm1789 Would this be enough to consider this story done? We have customizability for Tone, Temperature, and Max Words

[RobotGR]

General Report on the Feature For a general summary of each potential parameter that can be used on an API request please reference this link: https://platform.openai.com/docs/api-reference/chat/create

We have implemented 3 parameters. Tone, Temperature, and Max Words Further Parameters can be added in mern-template/src/client/ProfileDescription.tsx

This information is passed to mern-template/src/server/index.ts when you click "REWRITE WITH AI"

maxWords

• Currently on a slider system with a lower and upper bound that can be changed by the developer in ProfileDescription.tsx if needed
• This does not represent the minimum amount of words but the maximum. If limited information is given in the request, the API will not be able to create a very long description

tone

• Makes use of sending 2 messages to the API in a single request with different roles. The "system" role, allows us to give context to ChatGPT on how it is supposed to answer incoming questions. The "user" role is where we ask the specific questions that we want answered
• Each of these prompts can be changed depending on what you are trying to summarize or rewrite, as well as be used to control the mood of your writing

Temperature

• This can loosely be considered the creativity of the responses given by the ChatGPT API, which ranges from 0.0-2.0
• We have found that there are chances of highly random or non-sensical answers when Temperature > 1.5, but an increased level of Temperature increases the chances that the rewrite would be longer than the given information

The current Feature allows you to continuously see how changing parameters will affect the rewrite of your description which makes it very helpful for testing. There were hopes that word censorship could be added, which we believe could best be completed with further prompting specifications on the message request sent to the API. Our research has shown that asking the API to not use specific words can prove effective, but becomes less effective when the amount of words you are trying to censor and the length of your response increases. There is a specific parameter called logit_bias that excludes the usage of specific tokens (or parts of words that LLMs understand language through, like "have" would be split into ha and ve), but we believed that eliminating too many tokens would potentially limit the vocabulary of the responses (if we can't use the "ve" token because we want to try to eliminate the usage of delve, would it not eliminate the usage of have and other words ending in ve?). Both of these options could be explored if an app specifically requires censorship of specific words.

Troubleshooting Running the Code

• You need a .env file which contains the API Key
• make sure that you are on localhost:3000/ai-profile or localhost:8000/ai-profile (there is not currently a button that takes you from the main site to the ai-profile section)