TUNING CHATGPT MATHEMATICAL REASONING LIMITATIONS AND FAILURES WITH PROCESS SUPERVISION

Abstract: In ChatGPT, vast amounts of text-based data, generally scraped from the public internet, are used to train AI algorithms known as Large Language Models (LLMs). However, LLMs have some severe weaknesses, including the ability to make mistakes, "hallucinate" false information, and bias. They aren't magic, they're not artificial general intelligence, and they have these flaws. The tendency of the models to produce text that seems to be accurate but is actually untrue or not based on the input given is known as hallucination. For instance, even when a language model completely constructed the answer to a question concerning a historical occurrence that never took place, it may nonetheless produce a believable response. Hallucinations are the name for these fabricated LLM responses. Keywords: ChatGPT, public internet, Large Language Models (LLMs). Title: TUNING CHATGPT MATHEMATICAL REASONING LIMITATIONS AND FAILURES WITH PROCESS SUPERVISION Author: Indrasen Poola International Journal of Novel Research in Computer Science and Software Engineering ISSN 2394-7314 Vol. 10, Issue 2, May 2023 - August 2023 Page No: 55-66 Novelty Journals Website: www.noveltyjournals.com Published Date: 29-August-2023 DOI: https://doi.org/10.5281/zenodo.8296440 Paper Download Link (Source) https://www.paperpublications.org/upload/book/GENDER%20INEQUALITY-28082023-1.pdfInternational Journal of Novel Research in Computer Science and Software Engineering, ISSN 2394-7314, Novelty Journals, Website: www.noveltyjournals.co