Empowering the mass adoption of AI

A LLM is a language model consisting of a neural network with many parameters (typically billions of weights or more), trained on large quantities of unlabeled text using self-supervised learning or semi-supervised learning. They can be used for a variety of tasks such as text generation, text classification, question answering, and machine translation. Over time, these models have continued to improve, allowing for better accuracy and greater performance on a variety of tasks. We are now seeing the emergence of LLMs that can understand images, videos and audio as well (GPT-4o, Gemini 1.5 Pro, Claude 3.5 Sonnet).

At tokes compare, we allow you to compare open source and commercial LLMs and understand which one will meet your requirements. 

Open LLMs are AI-based models released under a license allowing free access, modification, and distribution (e.g., Llama 3). It is important to note that some open source LLMs may prohibit commercial use. Commercial LLMs, such as OpenAI's GPT-o1 or Claude Anthropic's 3.5 Sonnet, are developed by private entities that restrict access and usage, often requiring payment. The main differences between these models lie in their access, usage policies, and the availability of source code and training data.

The concept of a multimodal LLM represents a significant evolution in the field of Natural Language Processing (NLP) and Artificial Intelligence (AI). Traditional language models, such as the initial versions of GPT, were primarily text-based and relied on extensive text corpora for their training. The new LLMs such as GPT-o1 or Gemini are capable of multiple methods of input, such as photos, audio, and video, or what’s known as a multimodal model. The goal is to make the AI system more versatile and adaptable, allowing it to understand and generate information from various inputs beyond just text. By combining these different types of data, multimodal language models can achieve a more comprehensive understanding of the world, mirroring the way humans integrate information from different senses.

Tracking usage is essential for pay-as-you-go LLMs because it helps you monitor and manage your costs effectively. By understanding your usage patterns, you can optimize resource allocation and control spending on the LLM service. Additionally, monitoring usage can help you identify potential bottlenecks, optimize your application's performance, and ensure that you stay within the rate limits and usage allowances of your chosen pricing plan. Depending on usage, pay-as-you-go API pricing can be quite affordable compared to the subscription/purchasing plans but it is important to understand your usage. Some companies let customers set up an automatic notification to alert them when their credit is running short. You can choose the credit amount that will trigger the notification, and if it is reached at any point, you will be notified via email.

A token is a basic unit of text representation used in LLMs for processing and understanding the text.  Tokens are the smallest units of text that an LLM can understand and process. Tokens can be words, subwords, or characters, depending on the tokenization method employed.

When people refer to a Large Language Model, they are usually talking about a type of artificial intelligence model designed to understand, generate, and manipulate human language. These models are built using deep learning techniques and are trained on massive amounts of text data. As a result, they can comprehend and produce text in a coherent and contextually relevant manner, often resembling human-like language understanding.

Generative AI platforms, on the other hand, are systems that utilize these LLMs or similar generative models to create content. These platforms can generate text, images, music, or even code based on given inputs or prompts. They are capable of completing tasks like generating responses in a conversation, summarizing text, creating human-like articles, and more.

Large Language Models, like OpenAI's GPT series are often used as the foundation for generative AI platforms, enabling a wide range of applications such as chatbots, content generation, translation services, and virtual assistants - hence why it is good to know more about how they can help you transform your applications or business.

LLM APIs can be billed based on the volume of input they receive and the volume of output they generate (as well as the API flat rate call). Different providers measure these volumes differently, with some measuring tokens and others measuring characters (but the concepts remain the same). At tokes compare, we convert them all to a standard unit - 1 million (1/m) so you can see where you get the best value side-by-side.

Understanding the difference between a token and a character is essential to compare pricing across providers properly.

• A token is a unit of measure representing characters (see our blog section for more details on tokens and tokenization). A helpful rule of thumb is that one token generally corresponds to ~4 characters of text for common English text. This translates to roughly ¾ of a word (so 100 tokens ~= 75 words).

• A character is a single letter, number, or symbol. For example, the word “hello” has five characters and may be slightly larger than one token.

Retrieval-augmented generation (RAG) is a technique for enhancing the accuracy and reliability of generative AI models with facts fetched from external sources. With RAG, users can essentially have conversations with data repositories, opening up new kinds of experiences. Once companies get familiar with RAG, they can combine a variety of off-the-shelf or custom LLMs with internal or external knowledge bases to create a wide range of assistants that help their employees and customers.

In artificial intelligence, an agent is a computer program or system that is designed to perceive its environment, make decisions, and take actions to achieve a specific goal or set of goals. Technically, the new GPTs from OpenAI are agents. A Multi-Agent AI System consists of specialised agents interacting to achieve goals. It involves defining agents' roles and capabilities, and their interaction behaviors, including communication and decision-making processes, to efficiently achieve collective or individual objectives. This framework applies to autonomous vehicles, supply chains, smart grids, and customer service. Check out Autogen, crewai and Langraph to get started - these really are the future of AI.

GPUs, or Graphic Processing Units, are specialised circuits designed to speed up image, video, and animation creation, widely used in gaming and professional graphics. They're great for tasks that can run in parallel, making them ideal for machine learning and AI. The term "Language Processing Unit" (LPU), such as the new Groq LPU which can process token sequences much faster than GPUs, is hardware tailored for natural language tasks like parsing and generating human language.

Hallucinations occur when LLMs provide inaccurate or incomplete information, usually based on existing knowledge but applied incorrectly. Researchers demonstrate that hallucinations stem from the fundamental mathematical and logical structure of LLMs. It is, therefore, impossible to eliminate them through architectural improvements, dataset enhancements, or fact-checking mechanisms.

A knowledge graph, also known as a semantic network, represents a network of real-world entities—such as objects, events, situations or concepts—and illustrates the relationship between them. This information is usually stored in a graph database and visualized as a graph structure, prompting the term knowledge “graph.”