What is Large Language Model (LLM)? – TechTarget definition

Applications of AI

Large Language Models (LLMs) are a class of artificial intelligence (AI) algorithms that use deep learning techniques and large data sets to understand, summarize, generate, and predict new content.the term Generative AI It is also closely related to LLM. LLM is actually a type of generative AI specifically designed to help generate text-based content.

For thousands of years, humans have developed spoken language for communication. Language is at the core of all forms of human and technical communication. It gives you the words, semantics, and grammar you need to communicate your ideas and concepts. In the world of AI, language models serve a similar purpose, providing a foundation for communicating and generating new concepts.

The first AI language model traces its roots back to the early days of AI. The ELIZA language model debuted at his MIT in 1966 and is one of his earliest examples of AI language models. All language models are first trained on a set of data. It then uses various techniques to infer relationships and generate new content based on the trained data. Language models are commonly used in natural language processing (NLP) applications where users enter queries in natural language and produce results.

LLM is an evolution of the language model concept in AI, dramatically expanding the data used for training and inference. As a result, the capabilities of AI models are significantly improved. There is no universally accepted number for the size of the dataset required for training, but LLMs typically have at least a billion parameters or more. parameter A machine learning term for variables present in a trained model that can be used to infer new content.

Graph showing the number of GPT-4 parameters
As an LLM, GPT-4 overwhelms all its predecessors in terms of number of parameters.

The latest LLM appeared in 2017 and uses Transformer Neural Networks. transformerBy using a large number of parameters and transformation models, LLM can quickly understand and generate accurate responses, making AI technology broadly applicable to many different domains.

Some LLMs are foundation model, a term coined by the Stanford Institute for Human-Centered Artificial Intelligence in 2021. The underlying model is so large and influential that it serves as the basis for further optimizations and specific use cases.

How do large language models work?

LLM takes a complex approach involving multiple components.

At the base layer, LLM needs to be trained on large volumes. corpus — Usually petabytes of data in size. Training can be multi-step and usually starts with an unsupervised learning approach. In that approach, the model is trained on unstructured and unlabeled data. The advantage of training on unlabeled data is that much more data is often available. At this stage, the model begins to derive relationships between various words and concepts.

The next step for some LLMs is training and fine-tuning using a form of self-supervised learning. Here the data is labeled so that the model can more accurately identify different concepts.

LLM then performs deep learning as it passes through the Transformer Neural Network process. The Transformer architecture enables LLMs to use self-attention mechanisms to understand and recognize relationships and connections between words and concepts. That mechanism can assign a score, commonly called a score. weighta specific item (called token) to determine the relationship.

Once the LLM is trained, there exists a base on which AI can be used for practical purposes. By querying LLM with prompts, AI model inference can generate responses. This could be an answer to a question, newly generated text, summarized text, or sentiment analysis.

What are large language models used for?

LLM is becoming increasingly popular due to its broad applicability to various NLP tasks, such as:

  • text generation. The ability of LLM to generate text on the topic it was trained on is the primary use case.
  • translation. For LLMs trained in multiple languages, the ability to translate from one language to another is a common feature.
  • Summary of contents. Blocks of text or multi-page summaries are useful features of LLM.
  • Content rewriting. Rewriting part of the text is another function.
  • Classification and classification. LLMs can classify and classify content.
  • sentiment analysis. Most LLMs can be used for sentiment analysis, helping users better understand the intent of a piece of content or a particular response.
  • Conversational AI and chatbots. LLM enables conversations with users in a generally more natural way than previous generations of AI technology.

Most Common Uses of Conversational AI with chatbot, This can exist in various forms that users interact with in the query and response model. One of the most widely used LLM-based AI chatbots is ChatGPT, which is based on OpenAI’s GPT-3 model.

What are the advantages of large language models?

There are many benefits that LLM offers to organizations and users.

  • Scalability and adaptability. LLM serves as the foundation for customized use cases. With additional training in addition to LLM, you can create models that are fine-tuned to your organization’s specific needs.
  • Flexibility. A single LLM can be used for various tasks and deployments across organizations, users, and applications.
  • performance. Modern LLMs are typically high performance and can generate fast, low-latency responses.
  • Accuracy. As the number of parameters and the amount of data trained on LLM increases, the Transformer model can provide higher levels of accuracy.
  • Ease of training. Many LLMs are trained on unlabeled data, which helps speed up the training process.

What are the challenges and limitations of large language models?

Although there are many advantages to using LLM, there are also some challenges and limitations.

  • development cost. Run, LLM typically requires large amounts of expensive graphics processing unit hardware and large data sets.
  • operating costs. After a period of training and development, the cost of operating an LLM for the host organization can be prohibitive.
  • bias. A risk of AI trained on unlabeled data is bias. This is because it is not always clear whether known biases have been removed.
  • explainability. The ability to explain how LLM was able to produce a particular result is neither easy nor obvious to the user.
  • Hallucination. AI hallucinations occur when LLM provides inaccurate responses that are not based on trained data.
  • complicated. With billions of parameters, modern LLM is a very complex technology that can make troubleshooting particularly complex.
  • glitch token. A maliciously designed prompt that causes LLM to malfunction. glitch tokenare some of the emerging trends for 2022 and beyond.

What types of large language models are there?

A set of terms has evolved to describe different kinds of large-scale language models. Common types include:

  • Zero shot model. It is a large generalized model trained on a generic corpus of data that can give reasonably accurate results for common use cases without requiring additional training. GPT-3 is often considered a zero-shot model.
  • fine-tuned or domain-specific models. Additional training on top of zero-shot models like GPT-3 can create fine-tuned domain-specific models. One example is the OpenAI Codex, a domain-specific LLM for programming based on GPT-3.
  • language expression model. One example of a linguistic representation model is the Bidirectional Encoder Representation (BERT) from Transformers. It utilizes deep learning and transformers suitable for NLP.
  • multimodal model. Originally LLM was tuned exclusively for text, but its multimodal approach allows it to process both text and images. An example of this is GPT-4.

The future of large language models

The future of LLMs is still being written by the humans developing the technology, but there may also be futures that LLMs write themselves. The next generation of his LLMs will not have typical artificial intelligence or intelligence, but they will continually improve and become “smarter”.

LLM continues to be trained on ever-larger data sets, which are increasingly well filtered for accuracy and potential bias. Future LLMs may also do a better job than current generations in providing attribution and better explanations of how certain results were generated.

The graph shows how the AI ​​is biased

Enabling more precise information about domain-specific knowledge is another possible future direction for LLM. There is also a class of LLMs based on a concept known as knowledge search — Includes Google’s Retrieval-Augmented Language Model (REALM) — which enables training and inference on highly specific data corpora to enable today’s users to specifically search for content on a single site. It will be possible.

There is also work underway to optimize the overall size and training time needed for LLMs, such as Meta’s Large Language Model Meta AI (LLaMA), which is smaller than GPT-3, but proponents are more accurate. claims.

The future looks bright for LLM as technology continues to evolve in ways that increase human productivity.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *