Important terms of generative AI explained.
Today’s AI assistants provide answers to questions or perform simple, well-defined tasks. But they do not act independently. In addition, more complex tasks always require a human to act as a supervisor. AI agents, on the other hand, should find a solution on their own and pursue goals autonomously or semi-autonomously. AI agents use basic … Read more
A general world model is an ambitious concept in artificial intelligence. The goal is to create an AI system that can understand and simulate the world as comprehensively as a human. Imagine a virtual assistant that can not only generate text or images, but also understand, predict, and respond to complex real-world situations. For example, … Read more
Imagine you have a universal toolbox that contains many different tools, but is too large and cumbersome for certain tasks. To perform certain tasks efficiently, you can use small, specialized attachments called adapters. These adapters attach to the general-purpose tool and extend its function. For example, you can attach a screwdriver adapter to a drill … Read more
Today’s language models are primarily trained to provide helpful and easy-to-understand responses. At the same time, it is possible for the AI to make up information that fits the text perfectly and looks factual, but is actually made up. Such errors are often called hallucinations. These can be avoided, for example, by using an optimized … Read more
Alignment, in the context of artificial intelligence, means ensuring that AI systems are designed so that their goals and actions are consistent with the values and interests of humanity. An AI that is not properly aligned could pursue goals that are logical from its programming, but have a negative impact on humanity. This will be … Read more
Today, most AI applications run in the cloud on powerful, specialized computers in data centers. But experts say this will not always be the case. There will also be an increasing number of small models that run directly on users’ devices, from PCs to smartphones. This is called edge AI, after the concept of edge … Read more
Some AI applications are freely available. Examples include language models from French vendor Mistral or the Llama family from Facebook/Meta. However, it is not correct to call these “open source”. What you get as a user is the end result of the training, the core of a large language model called “weights”. At the same … Read more
Commercial AIs like ChatGPT or Claude have a mostly invisible system prompt that explains important rules and guidelines to the assistant. The system prompt therefore influences how an AI behaves and whether it refuses certain requests, for example. A member of the Claude team recently published their assistant’s system prompt on Twitter/X, explaining it line … Read more
Merging, in the context of generative AI, refers to the combination or fusion of different AI models or their characteristics. Similar to creating a collage, the best or desired features of multiple models are united into a new model. A practical example is the merging of different Stable Diffusion models, where one model’s ability to … Read more
Multimodal (from Latin: multi = many, multiple and modus = way, manner) refers, in the context of Artificial Intelligence, to an AI system’s ability to process and understand different types of input or “modalities” simultaneously. While earlier AI systems typically focused on a single form of communication – such as text or images – multimodal … Read more