Will JavaScript type annotations kill TypeScript?
The creators of Svelte and Turbo 8 both dropped TS recently saying that "it's not worth it".
Yes: If JavaScript gets type annotations then there's no reason for TypeScript to exist.
No: TypeScript remains the best language for structuring large enterprise applications.
TBD: The existing user base and its corpensource owner means that TypeScript isn’t likely to reach EOL without a putting up a fight.
I hope they both die. I mean, if you really need strong types in the browser then you could leverage WASM and use a real programming language.
I don’t know and I don’t care.
AI / Large Language Models

What Is a Large Language Model?

A primer on what large language models are, why they are used, the different types, and what the future may hold for LLM applications.
Sep 1st, 2023 5:00am by
Featued image for: What Is a Large Language Model?
Image via Pexels.

First things first. Let’s answer the question, “What does LLM stand for?” LLM stands for Large Language Model. Of course, that begs a very important second question, “What are large language models?” In this article, we will provide a large language model definition and discuss the LLM meaning. Use this resource to explore what large language models are, what LLMs are in the context of AI, why they are used, the different types of large language models, and what the future may hold.

LLM or Large Language Model

LLMs are becoming a major talking point among developers and data scientists who are keen to explore new ways to create advanced artificial intelligence (AI) projects that use deep learning techniques. Popular LLMs include OpenAI’s GPT, Google’s PaLM2 (which its chat product Bard is based on), and Falcon; with GPT, in particular, becoming a global phenomenon. As the topic becomes more popular, more and more people have become familiar with LLM standing for large language model.

What Is an LLM? Large Language Models Explained

Large language model definition: An LLM is a type of language model that is characterized by its large size, capable of incorporating billions of parameters to build complex artificial neural networks. These networks are powered by AI algorithms that employ deep learning techniques and use huge data sets to evaluate, normalize, and generate relevant content, as well as make accurate predictions. LLMs are often associated with generative AI, as they are typically designed to generate text-based content.

Compared to standard language models, LLMs process extremely large datasets — which can significantly increase the functionality and capabilities of an AI model. “Large” has no set definition, but typically large language models contain at least one billion parameters (machine learning variables).

LLMs are referred to as foundation models in natural language processing, as they are a single model that can perform any task within its remit. LLMs evolved from early AI models such as the ELIZA language model, first developed in 1966 at MIT in the United States. Present-day LLMs train on a set of data in their early stages and then develop using a range of techniques (training) to build relationships within the model and generate new content.

Natural language processing (NLP) applications commonly rely on language models, allowing users to input a query in natural language to generate a response.

Large Language Model Uses

What is an LLM used for? Like all AI systems, large language models are built to perform a function — often assisting with written and spoken language to help improve grammar or semantics, and generating ideas and concepts while conveying them in a way that is easy to understand.

LLMs can also be trained on code repositories that have been sourced from the internet, generating relevant snippets of code in a range of languages to assist developers and streamline the development process. Developers can simply enter a code-based prompt into an LLM, or a tool based on an LLM (such as GitHub Copilot), which will then generate usable code in the chosen programming language.

Why Use AI Large Language Models?

As AI large language models are not specific to an individual goal or task, they can be applied to almost any project. Referring back to ChatGPT, the LLM-based chatbot can generate a response for most queries, tapping into masses of data to deliver (mostly) factual, interesting, and even funny answers to a question. This vast potential is one of the core reasons LLMs are used.

Also, large language models do not need to be constantly refined or optimized, like standard models that are pre-trained. LLMs only require a prompt to perform a task, more often than not providing relevant solutions to the problem at hand.

However, despite the large number of benefits, LLMs have been known to suffer from hallucination problems. This refers to text generation that bears little or no relevance to the task, often containing inaccuracies and sometimes giving responses that don’t make sense or are far removed from real-world scenarios.

Common large language model uses and LLM projects include:

  • LLMs can be trained in a range of languages to quickly translate one from the other. Falcon is an LLM that provides this functionality.
  • Bard and ChatGPT are examples of popular text-generation tools that use large language models. These LLMs can rewrite a block of text to improve it grammatically or give it a different style or tone of voice. They also can categorize and classify content to make it easier to understand.
  • The LLMs referred to above can also summarize content from large blocks of text or multiple pages to assist users in their research. Text can then be analyzed for sentiment to help users understand its overall intent: it is particularly useful for educational and learning purposes.
  • LLMs are being used to create better conversational chatbots that generate more natural, useful, and insightful responses. This allows users to converse about anything that comes to mind with minimal restrictions.
  • LLM models can streamline and speed up the software development process, generating snippets of code in a chosen programming language based on a developer’s prompt.

The Different Types of Large Language Models

Below is a summary of four different types of large language models that you are likely to encounter.

1. Zero Shot

A zero-shot model is a standard LLM meaning that it is trained on generic data to provide results for general use cases to a certain degree of accuracy. These models do not require any additional training.

2. Fine Tuned or Domain Specific

Fine-tuned models receive additional training to expand on the initial zero-shot model to improve its effectiveness. OpenAI Codex is an example of this and is often used as an auto-completion programming tool for projects based on GPT-3.

3. Language Representation

Language representational models use deep learning techniques and transformers (the architecture that gave rise to generative AI) that are suitable for natural language processing. This enables languages to be converted into a visual medium, such as writing.

4. Multimodal

Multimodal LMMs can handle both text and images, unlike older LMMs which were designed to generate just text. An example is GPT-4, the newer, multimodal version of GPT.

Large Language Models vs. Other Machine Learning Models

To determine when it is viable to use a large language model instead of other machine learning models, it is important to establish the advantages and limitations of LLMs when compared to models that use smaller data sets.

Advantages of LLMs

  • Models can be fine-tuned for a specific purpose with additional training.
  • LLMs can perform multiple tasks and be served for a range of deployments.
  • These models can easily be trained on unlabeled data.
  • LMMs generate rapid responses with low latency.
  • The large number of parameters and level of training data mean LLMs have access to a much wider knowledge base compared to standard models, making them capable of generating much more in-depth and sophisticated responses.

Limitations of LLMs

  • Development costs can be high due to the need for expensive hardware.
  • LLMs can have high operating costs.
  • LLMs are extremely complicated due to the billions of parameters involved.
  • In some cases, it is difficult to ascertain why an LLM has generated a result.
  • LLMs can be subject to glitch tokens, malicious prompts that cause a malfunction.
  • Models trained on unlabeled data can possess a certain level of bias.
  • LLMs can sometimes produce hallucinations, which are inaccurate responses.


So, what is a large language model? In reality, it can be so many things as the potential of large language models is vast. These models have the ability to revolutionize various domains, from natural language processing to text generation. However, it is important to note that the true potential of these models is ultimately shaped by the humans who develop and utilize them.

While the idea of artificial intelligence, machine learning and large language models evolving into sentient programs akin to those portrayed in science fiction movies may be purely speculative, their impact on our society and industries will undoubtedly continue to grow.

Industries that are sure to benefit from this projected change are tech, healthcare, gaming, finance, and robotics — with more advanced modalities expanding the use cases of LLMs, as well. Evolving from the standard text-to-text or text-to-image responses, with text-to-3D and text-to-video now possible.

This could see LLMs being used to design complex blueprints for robotic systems, or generate 3D characters and environments in video games. Meanwhile, advancements in digital biology could help design models that could predict changes in a human’s body, revolutionizing scientific research in the health sector.

As researchers and engineers push the boundaries of these technologies, we can expect to see these and more fascinating advancements and applications arise.

Group Created with Sketch.
THE NEW STACK UPDATE A newsletter digest of the week’s most important stories & analyses.