The critique of "LLM slop" originates from observations of AI-generated responses, often used in search engines or chatbots, where large language models generate expansive, verbose responses that can sometimes overwhelm users or offer redundant information. Despite advancements in natural language understanding, LLMs sometimes produce answers that are overly descriptive, fail to get to the point, or provide redundant information. This is a limitation attributed to the way LLMs attempt to cover a wide range of possible user intentions or questions, sometimes leading to more verbose responses than desired.
The rise of "LLM slop" coincided with the mainstream adoption of AI models like OpenAI's GPT-3 (released in 2020) and Google’s LaMDA, as these models brought natural language generation to the forefront of web applications. The term became popular in tech discourse around 2023-2024, reflecting growing frustrations with the verbose nature of AI responses in commercial applications.
Major companies such as OpenAI and Google have driven the development of LLMs, contributing significantly to the current landscape of natural language generation. These companies' models are often the source of AI-generated content criticized as "LLM slop."