Joanna Baran

Jan Kocoń, Igor Cichecki, Oliwier Kaszyca et al.

OpenAI has released the Chat Generative Pre-trained Transformer (ChatGPT) and revolutionized the approach in artificial intelligence to human-model interaction. Several publications on ChatGPT evaluation test its effectiveness on well-known natural language processing (NLP) tasks. However, the existing studies are mostly non-automated and tested on a very limited scale. In this work, we examined ChatGPT's capabilities on 25 diverse analytical NLP tasks, most of them subjective even to humans, such as sentiment analysis, emotion recognition, offensiveness, and stance detection. In contrast, the other tasks require more objective reasoning like word sense disambiguation, linguistic acceptability, and question answering. We also evaluated GPT-4 model on five selected subsets of NLP tasks. We automated ChatGPT and GPT-4 prompting process and analyzed more than 49k responses. Our comparison of its results with available State-of-the-Art (SOTA) solutions showed that the average loss in quality of the ChatGPT model was about 25% for zero-shot and few-shot evaluation. For GPT-4 model, a loss for semantic tasks is significantly lower than for ChatGPT. We showed that the more difficult the task (lower SOTA performance), the higher the ChatGPT loss. It especially refers to pragmatic NLP problems like emotion recognition. We also tested the ability to personalize ChatGPT responses for selected subjective tasks via Random Contextual Few-Shot Personalization, and we obtained significantly better user-based predictions. Additional qualitative analysis revealed a ChatGPT bias, most likely due to the rules imposed on human trainers by OpenAI. Our results provide the basis for a fundamental discussion of whether the high quality of recent predictive NLP models can indicate a tool's usefulness to society and how the learning and validation procedures for such systems should be established.

Joanna Baran, Michał Kajstura, Maciej Ziółkowski et al.

Every day, the world is flooded by millions of messages and statements posted on Twitter or Facebook. Social media platforms try to protect users' personal data, but there still is a real risk of misuse, including elections manipulation. Did you know, that only 13 posts addressing important or controversial topics for society are enough to predict one's political affiliation with a 0.85 F1-score? To examine this phenomenon, we created a novel universal method of semi-automated political leaning discovery. It relies on a heuristical data annotation procedure, which was evaluated to achieve 0.95 agreement with human annotators (counted as an accuracy metric). We also present POLiTweets - the first publicly open Polish dataset for political affiliation discovery in a multi-party setup, consisting of over 147k tweets from almost 10k Polish-writing users annotated heuristically and almost 40k tweets from 166 users annotated manually as a test set. We used our data to study the aspects of domain shift in the context of topics and the type of content writers - ordinary citizens vs. professional politicians.