Large Language Model: Difference between revisions
Aishavingfun (talk | contribs) No edit summary |
Aishavingfun (talk | contribs) mNo edit summary |
||
| Line 9: | Line 9: | ||
== Applications == | == Applications == | ||
LLMs can write, code, summarize, and generally process natural language textual input in a credible fashion to improve productivity and assist in creative endeavors. They have been used to automate common support tasks, sift and classify medical data, summarize complex reports for distribution and far more. | LLMs can write, code, summarize, and generally process natural language textual input in a credible fashion to improve productivity and assist in creative endeavors. They have been used to automate common support tasks, sift and classify medical data, summarize complex reports for distribution and far more. One must always, however, be aware of [[bias]] when using such models, as they are only as good as the data they were trained with. | ||
Latest revision as of 14:24, 11 April 2024
Large Language Models are of a type of Artificial Intelligence trained using deep learning and massive text-based data set to generate textual outputs via natural language input from the user.
Like Stable Diffusion, and other image generation AI platforms, they are considered a type of Generative AI. Popularized by services such as OpenAI's ChatGPT, most major technology players now offer the services of an LLM to their users and customers.
Underlying Technology
LLMs are a type of neural network, built with a decoder-only transformer-based architecture. Data (text, for LLM) is broken down into tokens, which are then converted to numerical vectors. Those are used in a probabilistic model to come up with a coherent output, based on the quality of the input data.
Applications
LLMs can write, code, summarize, and generally process natural language textual input in a credible fashion to improve productivity and assist in creative endeavors. They have been used to automate common support tasks, sift and classify medical data, summarize complex reports for distribution and far more. One must always, however, be aware of bias when using such models, as they are only as good as the data they were trained with.