Skip to Main Content

Large Language Models: ChatGPT and Beyond

A guide to the various artificial intelligence (AI) algorithms that use deep learning techniques and massively large data sets to understand, summarize, generate and predict new content.

Science & Engineering Librarian

Profile Photo
Denise Wetzel
206 Davey Lab
Physical & Mathematical Sciences Library
University Park

What are LLMs?

Large Language Models (LLMs) refer to large general-purpose language models that can be pre-trained and then fine-tuned for specific purposes. They are trained to solve common language problems, such as text classification, question answering, document summaries, and text generation. The models can then be adapted to solve specific problems in different fields using a relatively small size of field datasets via fine-tuning.

The ability of LLMs taking the knowledge learnt from one task and applying it to another task is enabled by transfer learning. LLMs predict the probabilities of next word (token), given an input string of text, based on the language in the training data. Besides, instruction tuned language models predict a response to the instructions given in the input. These instructions can be "summarize a text", "generate a poem in the style of X", or "give a list of keywords based on semantic similarity for X".

LLMs are large, not only because of their large size of training data, but also their large number of parameters. They display different behaviors from smaller models and have important implications for those who develop and use A.I. systems. To develop effective LLMs, researchers must address complex engineering issues and work alongside engineers or have engineering expertise themselves.

Definitions to Know

Artificial intelligence, is "a branch of computer science dealing with the simulation of intelligent behavior in computers."

Machine Learning is artificial intelligence running "computer programs that learn from examples and from experience."

Narrow AI is different than General AI because it "is created to solve one given problem." This kind of AI is sometimes labeled "Weak AI."

General AI, or Artificial General Intelligence (AGI), is "a more advanced form of artificial intelligence that can learn and adapt to its environment." Many science-fiction AI are considered AGI such as J.A.R.V.I.S. in Iron Man; or the character Cyberdyne Systems Model 101 in The Terminator.

Natural Language Processing, or NLP, is a branch of AI allowing computers to understand text and speech in the same way humans do.  Baumann & Schuler (2023) specify that "natural language processing tasks range from text searches (such as web searches) to interaction with spoken language (such as with Siri, Alexa, or similar voice-controlled agents)."

Large Language Models (LLM) are the technology underlying generative AI. Tech writer Bob Sharp's article describes what these models are, how they are created and how they work, and he lists some of their limitations, too. Wired's David Nield's article discusses large language models and how they work.

Reinforcement Learning on Human Feedback (RLHF) is the activity occurring within AI-generated text and images when human beings, in the roles of supervisors and end users, comment on AI output to help it become more accurate.

Neural Networks mimic the work of the human brain's neuron network. Neural networks in AI "learn to perform better by consuming input, passing it up through the ranks of neurons, and then comparing its final output against known results, which are then fed backwards through the system to alter how the nodes perform their computations."

Deep Learning is another sub-field of machine learning using neural networks to complete its tasks, by "processing multiple layers of programmed algorithms ... and then learning complicated concepts ... through experience." Google's search-engine algorithm uses neural-network processing.  

A Data Corpus is a collection "of data on which it (generative AI) is trained, then [it] respond[s] to prompts with something that falls within the realm of probability as determined by that corpus." This corpus usually is large in size and contains many kinds of files; in order for a corpus to be the best possible, it should be high-quality, vast, clean, and without biases.

The Black Box Problem is the situation that arises when an AI has done so much work on its own that its developers no longer understand how it is making its decisions and predictions. Heller (2021) says, "For any AI decision that has an impact - not only a life-and-death impact, but also a financial impact or a regulatory impact -- it is important to be able to clarify what factors went into the model's decision."

AI Hallucinations are items created by an AI model that "generates output that deviates from what would be considered normal or expected based on the training data it has seen."

Interesting Reads related to LLMs