Breaking the Bank with ChatGPT: Few-Shot Text Classification for Finance

We propose the use of conversational GPT models for easy and quick few-shot text classification in the financial domain using the Banking77 dataset. Our approach involves in-context learning with GPT-3.5 and GPT-4, which minimizes the technical expertise required and eliminates the need for expensive GPU computing while yielding quick and accurate results. Additionally, we fine-tune other pre-trained, masked language models with SetFit, a recent contrastive learning technique, to achieve state-of-the-art results both in full-data and few-shot settings. Our findings show that querying GPT-3.5 and GPT-4 can outperform fine-tuned, non-generative models even with fewer examples. However, subscription fees associated with these solutions may be considered costly for small organizations. Lastly, we find that generative models perform better on the given task when shown representative samples selected by a human expert rather than when shown random ones. We conclude that a) our proposed methods offer a practical solution for few-shot tasks in datasets with limited label availability, and b) our state-of-the-art results can inspire future work in the area.

Path planning with Inventory-driven Jump-Point-Search

In many navigational domains the traversability of cells is conditioned on the path taken. This is often the case in video-games, in which a character may need to acquire a certain object (i.e., a key or a flying suit) to be able to traverse specific locations (e.g., doors or high walls). In order for non-player characters to handle such scenarios we present invJPS, an "inventory-driven" pathfinding approach based on the highly successful grid-based Jump-Point-Search (JPS) algorithm. We show, formally and experimentally, that the invJPS preserves JPS's optimality guarantees and its symmetry breaking advantages in inventory-based variants of game maps.

Action-based Character AI in Video-games with CogBots Architecture: A Preliminary Report

In this paper we propose an architecture for specifying the interaction of non-player characters (NPCs) in the game-world in a way that abstracts common tasks in four main conceptual components, namely perception, deliberation, control, action. We argue that this architecture, inspired by AI research on autonomous agents and robots, can offer a number of benefits in the form of abstraction, modularity, re-usability and higher degrees of personalization for the behavior of each NPC. We also show how this architecture can be used to tackle a simple scenario related to the navigation of NPCs under incomplete information about the obstacles that may obstruct the various way-points in the game, in a simple and effective way.