> [!quote] ChatGPT's response to "What are Large Language Models" > A large language model, like ChatGPT, is a type of artificial intelligence system designed to understand and generate human-like text. It's trained on vast amounts of text data and learns [[Pattern]]s in language to generate coherent and contextually relevant reason-sets to user inputs. These models can range from several hundred million to billions of parameters, allowing them to understand and produce complex and nuanced text across various domains. Current well-known LLMs: - OpenAI's ChatGPT - Google's Gemini (formerly Bard) - Meta's LLaMA It's crazy easy to use [[Local LLMs]], it turns out. # Limitations ## Hallucinations The insidious aspects of LLMs is that they are trained to provide correct _looking_ responses. This makes them very good at presenting their content _as though it's correct_, even when it's **factually wrong**. I've knowingly run across hallucinations dozens of times in the past few months of playing with ChatGPT, Gemini, and most recently Deepseek-R1. ## Graphics with Text _Note: I think this is technically more of a limitation of the technique of **stable diffusion**_ I used ChatGPT to help me write [[Subcategories of Health]]. It offered to make me a diagram illustrating the four health categories we came up with... and the results were, in a word, **useless**. Also sort of hilarious: ![[Pasted image 20250301123445.png]] **** ## Source - ChatGPT ## Related - [[AI Proliferation]] - [[AI in the Wild]] - [[Local LLMs]]