Artificial Intelligence, Blog
AI trends 2024
![](https://www.datahouse.ch/wp-content/uploads/jan-winkler-profile-pic--48x48.jpg)
22 Jan 24
The world of artificial intelligence (AI) is evolving dynamically, producing groundbreaking innovations every year. Below we present three AI trends for 2024 that have the potential to further change the way we interact with AI technology. These trends include the use of Retrieval Augmented Generation (RAG) systems in a business context, the increasing importance of multimodal AI and the miniaturization of AI models for their use on smartphones.
Trend 1: RAG and integration of business-specific data
What does RAG mean?
Retrieval Augmented Generation (RAG) marks an innovative advance in Artificial Intelligence (AI), particularly highlighted by its utility in the business world. Unlike traditional AI-based chat systems, RAG systems combine text generation with the ability to extract relevant information from external data sources. They use a Large Language Model (LLM) trained on a variety of texts to understand complex queries and generate responses in natural language. This technology enables RAG systems to incorporate the latest data and insights, which is particularly beneficial in knowledge areas such as sustainability in the real estate or financial sectors.
Effective use of business data in RAG systems
The strength of RAG systems lies in their ability to use both structured and unstructured data effectively. They can extract specific information from structured data sources such as databases or spreadsheets while processing unstructured data from reports, emails, or web content. This versatile processing capacity enables RAG systems to draw on a wide range of business data for informed decision-making and analysis. For example, a RAG system trained on existing text can incorporate new data on energy consumption and air quality to provide accurate recommendations for green building projects or investments.
Trend 2: Multimodality of models
What does multimodality mean?
Multimodal models are a type of AI that can process and understand different types of data such as text, images, and audio. They are designed to integrate information from these different data types to perform complex tasks, such as generating images from text descriptions or answering questions based on a combination of text and images. This allows them to handle a wider range of inputs and applications than models that are limited to a single data type.
What are the benefits of multimodal models?
For users, these types of models can be very helpful as they can cover different requirements in a single application. What models such as Gemini (Google) or GPT 4.0 (openAI) can already do today, for example, is to recognize mathematical questions on images and provide users with a useful answer:
The image on the left shows the “prompt” – i.e. the problem. The prompt shows a solution to a physics problem. On the right you can see the model’s answer and how the model (here Gemini) argues. The model recognizes that the problem has been solved incorrectly and shows a correct solution.
Trend 3: AI models are getting smaller, and applications are moving towards smartphones
Transition to smartphones
The continuous development of artificial intelligence is leading to a significant trend: the adaptation and miniaturization of AI models for use on smartphones. At a time when smartphones play a central role in the organization of many people’s private and business lives, this development is becoming increasingly important. The use of AI models on smartphones not only improves user-friendliness, but also extends the functionality of a wide range of applications, such as AI-supported image and video editing or assistance systems. Another important aspect of this development is the improvement in privacy: thanks to powerful, locally operated AI models, user data no longer needs to be transferred to the cloud. Instead, data can be processed directly on the smartphone by AI applications, which minimizes the risk of data leaks and increases user privacy.
The challenge of adapting to smartphones
Adapting complex AI models, such as GPT-4, to smartphones is a technical challenge. These models, which currently run on powerful servers, must be significantly scaled down for use on the limited resources of a smartphone. This affects both the memory size and the processing capacity. Techniques such as model pruning, in which unimportant parts of the neural networks are removed to make them more efficient, and knowledge distillation could help to reduce the size of the models without significantly impairing their performance. Another current trend is the advancement of smartphone chips, including the integration of specialized neural processing units (NPUs) specifically designed to perform AI tasks efficiently.
These trends indicate that AI technologies are becoming increasingly present in our daily lives and in the world of work. As these systems continue to evolve, a picture is emerging in which their application becomes more seamless, intuitive, and efficient.
share