Artificial intelligence is a deep and complex world. Scientists working in this field often rely on jargon and terminology to explain what they are working on. As a result, these technical terms are often necessary in reporting in the artificial intelligence industry. So we thought it would be helpful to put together a glossary containing definitions of the most important words and phrases to use in our article.
Researchers will continuously add new entries and regularly update this glossary to add new entries, revealing new ways to boost the frontier of artificial intelligence while identifying new safety risks.
AI agents refer to tools that use AI technology to perform a set of tasks beyond what a more basic AI chatbot can do. However, as I explained before, there are many moving pieces in this emergency space. So referencing AI agents can mean different people are different. In addition, the infrastructure is still built to provide the expected functionality. However, the basic concept refers to an autonomous system that utilizes multiple AI systems to perform multi-step tasks.
Given a simple question, the human brain can answer it without much thinking about it – “Which animal is taller, between a giraffe or a cat?” However, there are often mediation procedures, so you often need a pen and paper to come up with the correct answer. For example, if a farmer has a chicken and a cow, and together they have 40 and 120 legs, you might need to write down a simple equation to come up with an answer (20 chickens and 20 cows).
In the AI context, inference of a chain of large language models means decompose the problem into smaller, intermediate steps to improve the quality of the final result. It usually takes time to get an answer, but the answer is more likely to be correct, especially in the context of logic or coding. The so-called inference models have been developed from traditional large-scale language models and are optimized for chain thinking thanks to reinforcement learning.
(Reference: Large-scale language model)
A subset of self-improvement machine learning, in which AI algorithms are designed with multilayered artificial neural network (ANN) structures. This allows for more complex correlations to be created compared to simpler machine learning-based systems, such as linear models and decision trees. The structure of deep learning algorithms draws inspiration from the interconnected pathways of neurons in the human brain.
Rather than requiring human engineers to define these features, deep learning AI can identify important characteristics of the data itself. This structure also supports algorithms that can be learned from errors, improving its own output through the process of iteration and adjustment. However, deep learning systems require many data points (over millions) to achieve good results. It also usually takes time to train deep learning and simpler machine learning algorithms. Therefore, development costs tend to be high.
(Reference: Neural Network)
This implies further training of AI models aimed at optimizing performance for more specific tasks or areas than previously the focus of training. It is usually about supplying new specialized (i.e. task-oriented) data.
Many AI startups have acquired large-scale language models as starting points for building commercial products, but are competing to increase the utility of the target sector or task by supplementing previous training cycles with fine tuning based on domain-specific knowledge and expertise.
(Reference: Major Language Models (LLM))
A large language model, or LLM, is an AI model used by popular AI assistants such as ChatGpt, Claude, Google’s Gemini, Meta’s Ai Llama, Microsoft Copilot, or Mistral’s LE Chat. When you chat with your AI assistant, you interact with the help of a large language model that directly handles requests or a variety of available tools, such as web browsing and code interpreting.
AI Assistant and LLM can have different names. For example, GPT is Openai’s large-scale language model, and ChatGPT is an AI assistant product.
LLM is a deep neural network made up of billions of numerical parameters (or see weights) that learn the relationships between words and phrases and create representations of languages that are multidimensional maps of words.
These are created by encoding patterns in billions of books, articles, and transcripts. When you prompt LLM, the model generates the most likely pattern to fit the prompt. Next, based on what was said before, we evaluate the next word that is most likely after the last word. Repeat, repeat, repeat.
(Reference: Neural Network)
A neural network refers to a multi-layered algorithm structure that supports deep learning. Furthermore, it refers to the entire boom in generation AI tools following the emergence of wider, larger language models.
The idea of taking inspiration from the tightly interconnected pathways of the human brain as a design structure for data processing algorithms dates back to the 1940s, but it was the much more recent rise of graphical processing hardware (GPUs) through the video game industry, and truly unleashed the power of theory. These chips have proven suitable for training algorithms with more layers than previous epochs can. This allows for far better performance in many domains, such as speech recognition, autonomous navigation, or drug discovery.
(Reference: Major Language Models (LLM))
Weights is the heart of AI training because it determines how important (or weight) is given to the various features (or input variables) of the data used to train the system. This forms the output of the AI model.
Put another way, weights are numerical parameters that define what is most prominent in the dataset of a given training task. It accomplishes the function by applying multiplication to the input. Model training usually starts with randomly assigned weights, but as the process unfolds, the weights are adjusted as the model tries to reach an output that matches the target more closely.
For example, AI models for predicting home prices trained with historical real estate data for target locations include functionality weights such as the number of bedrooms and bathrooms, whether the site is separated, semi-isolated, there is a parking lot, or no garage.
Ultimately, the weights that the model adheres to each of these inputs reflect how much of the property’s value, based on the specified dataset.
Source link
