Year: 2025
Type: article
Abstract: Now that we live in a world where most of the text in recent online interactions that we come across seems to be generated by LLMs, it becomes critical to understand the nature of statements being generated by LLMs. Technology has always been sold to humans under the tag that it is foolproof and will make lives easier. LLMs produce text by predicting the next token or sequence based on probabilities derived from their training data. A question then arises, whether they generate a ‘probability statement’ or ‘probability of a statement’. The difference between the two may seem elusive, but it is actually quite obvious. This paper intends to bring forward that difference to its audience, who, in turn, can understand the capabilities of the machine they are using and adapt a better framework to judge and use the response generated by LLM models in their applications.
Author: Reeshabh Choudhary
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W4415616278