‘These models will always hallucinate’: Seth Dobrin on LLMsDek

Limitations of Large Language Models in Financial Services

Large language models (LLMs) present several challenges in the financial sector, particularly concerning their reliability. These models, built on transformer architecture, have evolved from their original purpose of language translation to handling a wide array of tasks, leading to increased instances of 'hallucinations'—incorrect or fabricated information. Despite banks experimenting with LLMs for tasks like automating customer interactions and generating summaries, experts emphasize that these models are inherently prone to errors, especially when critical decisions are at stake. As a result, their use should be cautiously approached, especially in contexts requiring high accuracy and responsibility.