This repository contains simple Python library created by me and Dmitry during "Mathematics in AI" summer camp by Yandex Education which took place in Innopolis University. The idea behind it is that you input a prompt like "You're happy" and get some nice plots as an output. It delivers 4 predefined psychological tests (all in Russian) and a basic agent for interacting with YandexGPT while also providing some level of extensibility.
Human decisions are often strongly influenced by emotions, whereas LLM ones are strictly logical (not that logical, but logical enough). Although LLMs are able to understand emotions to some extent, it's hard to evaluate their ability to express them. So, during the camp, we chose to create a project around measuring how "emotionally aligned" different models are, based on different types of prompts they were given.
Sure! You can find plots generated by demo.py (6 emotions with 6 ways to define each, all across 2 versions of YandexGPT) in this Markdown file (in Russian). Unfortunately, we decided to keep private both prompts (due to the fact that they were created by other team) and LLM answers.