Aleph Alpha introduces new generation of AI language models

Aleph Alpha, an AI startup based in Heidelberg, Germany, has presented its new Control Model generation. The models are specifically designed to improve natural language processing and solve computational linguistic tasks.

Heidelberg - From now on, partners and customers can use the control variant for each model. They are ideal for applications such as chatbots and digital assistants. They are particularly suitable for applications such as chatbots and digital assistants. A notable milestone of these new models is that they integrate important source verification and explainability features for AI-generated content.

Our Control models are capable of delivering significantly better answers - in pilot customer evaluations, they were neck-and-neck with the best models on the market

Jonas Andrulis, CEO & Founder Aleph Alpha

A key difference is that the models move from "few-shot prompting" to "zero-shot prompting." Typically, large language models are trained with Few-Shot Prompting, where the model is given a series of examples to develop context structure. In contrast, control models can generate more efficient and robust responses for various tasks by requiring only a single instruction - this is called zero-shot prompting. This shortens the development cycles of chatbots and digital AI assistants. Users benefit from natural and seamless interactions between humans and machines, as the models provide improved instructional understanding and instant context.

Transparency in AI

All control models can understand relationships in information and display the textual sources that led to or contradict the system's generated responses based on verified facts. This feature was introduced by Aleph Alpha in April 2023 and open-sourced in an academic publication. It enables transparency and traceability of AI-generated content, which meets the upcoming EU AI Act regulatory requirements for explainability and traceability. According to Aleph Alpha spokesperson Andrulis, their explainability not only provides the necessary context for the models' factual statements, but also enables verification and traceability of complex and critical AI support.

We first reported on the EUR23 million funding round for Aleph Alpha in 2021, and then in May 2023 we reported on the investment in F13, the artificial intelligence for administration in Baden-Württemberg.