The ETH Zurich researchers have developed a method that makes AI answers more reliable over time. Their algorithm is highly selective in choosing data. Additionally, up to 40 times smaller AI models can achieve the same output performance as the best large AI models.

ChatGPT and similar tools often amaze us with the accuracy of their answers, but also often lead to doubt. One of the big challenges of powerful AI response machines is that they serve us with perfect answers and obvious nonsense with the same ease. One of the major challenges is how the underlying large language models (LLMs) of AI deal with uncertainty. It has been very difficult until now to judge whether LLMs focused on text processing and generation generate their answers on a solid foundation of data or whether they are on uncertain ground.

Researchers from the Institute for Machine Learning at the Department of Computer Science at ETH Zurich have now developed a method to specifically reduce the uncertainty of AI. 'Our algorithm can specifically enrich the general language model of AI with additional data from the relevant thematic area of the question. In combination with the specific question, we can then specifically retrieve those relationships from the depths of the model and from the enrichment data that are likely to generate a correct answer,' explains Jonas Hübotter from the Learning & Adaptive Systems Group, who developed the new method as part of his PhD studies.

➀ The Q1 UK connector market saw a strong performance with sales up 11% over Q4 2024 and 7% on Q1 2024.

➁ Mass Transport and Data Processing were the standout sectors, with significant growth.

➂ The order intake in Q1 2025 was the second highest since Q2 2022, but there were significant variations in market performance across different sectors.

➀ The Fraunhofer IPMS is involved in a research project called InSeKT to develop new technological approaches for integrating AI at the edges of IT networks.

➁ The project aims to enable complex calculations directly where data is generated, improving data protection and real-time capabilities.

➂ Fraunhofer IPMS is working on sensor technology, including gas analysis using IMS, near-infrared photodetector evaluation, and adapted use of CMUTs for improved imaging.

The Fraunhofer Institute for Photonics Microsystems (IPMS) is involved in an interdisciplinary research project called 'InSeKT' (Development of Intelligent Sensor Edge Technologies). This project, carried out by the Technical University of Wildau, the Leibniz Institute for Innovative Microelectronics (IHP), and the Fraunhofer IPMS, aims to integrate artificial intelligence (AI) more effectively at the 'edges' of IT networks. The project focuses on miniaturized sensor structures and the integration of electronic components, with the goal of enabling complex calculations directly at the data source, such as at the sensor itself.

Current data processing with AI often occurs through central cloud computing solutions, leading to data transfer over large distances and potential data leaks. The project addresses this by promoting decentralized data processing for improved data security and real-time system capabilities.

The project covers various areas, including gas analysis using ion mobility spectrometers (IMS), data-supported evaluation of photodetectors for the near-infrared wavelength range, and the adapted use of capacitive microelectromechanical ultrasonic transducers (CMUTs) for improved imaging. The generated data will be used to train Edge-KI systems for fast and accurate data processing.

➀ The University of Ulm and Duisburg-Essen, along with Bosch and Nokia, organized a demo event for connected cooperative driving as part of the EU PoDIUM project.

➁ A live demonstration showcased the seamless coordination of automated vehicles in complex traffic situations.

➂ The research focuses on enabling secure interaction between automated vehicles using sensor data and a cooperative maneuver planner.

➀ Retrieval Augmented Generation (RAG) is being developed by Fraunhofer IWU to streamline the process of finding crucial information in extensive technical and legal texts.

➁ The technology is designed to complement Large Language Models (LLMs) by providing precise and exhaustive information.

➂ The team at IWU is using the EU Machinery Regulation (2023/1230) as a demonstration of the technology's capabilities.

➀ Information Theory has transitioned from mathematical textbooks to engineers' workbenches.

➁ The 4th London Symposium on Information Theory highlighted its practical applications.

➂ The UK delegation focused on the interaction of the theory with human activities like learning and speech.

The Fraunhofer Institute for Integrated Circuits IIS has developed an AI chip for processing Spiking Neural Networks (SNNs). The SENNA spiking neural network inference accelerator, inspired by brain function, consists of artificial neurons and can process electrical impulses (spikes) directly. Its speed, energy efficiency, and compact design enable the use of SNNs directly where data is generated: in edge devices.

SNNs consist of a network of artificial neurons connected by synapses. Information is transmitted and processed in the form of electrical impulses, allowing pulsing networks to be the next step in artificial intelligence: faster, more energy-efficient, and closer to the processing method of the human brain. To bring these advantages into application, small, efficient hardware that mimics a structure of neurons and synapses is needed. For this, the Fraunhofer IIS has developed the neuromorphic SNN accelerator SENNA as part of the Fraunhofer project SEC-Learn.

SENNA is a neuromorphic chip for fast processing of low-dimensional time series data in AI applications. The current version consists of 1024 artificial neurons on less than 11 mm² of chip area. Its low reaction time down to 20 nanoseconds ensures precise timing in time-critical applications at the edge. This makes it particularly strong in real-time event-based sensor data processing and in closed control systems, such as the control of small electric motors with AI. With SENNA, AI-optimized data transmission can be realized in communication systems. There, the AI processor can analyze signal streams and adjust transmission and reception methods as needed to improve efficiency and performance.

➀ The EU's ODISSEE project aims to find ways to process the volume of scientific data produced by research infrastructures using AI data processing.

➁ The project, funded by the European Union, targets fundamental mysteries such as the nature of dark matter.

➂ Research bodies like CERN's HL-LHC and SKAO will generate massive amounts of raw data, which current digital technologies cannot handle.

➀ Computers began to sing 63 years ago with IBM's 7094 being the first warbling data cruncher;

➁ The 7094 sang a tune;

➂ The article discusses the significance of this event in the history of computers.