Hamburg tests homegrown generative AI

The German city of Hamburg is trialling a large language model (LLM) named LLMoin.

LLMs are a type of generative artificial intelligence focused on text-based content.

The pilot aims to show how LLMs could be used within Hamburg’s public administration and which functions are useful to staff. The city also wants to understand the technical and organisational requirements that are necessary to make the technology available to a large number of employees in the future.

“The aim is to gain insights for a medium-term LLM strategy for the City of Hamburg,” a statement said.

The LLMoin tool is an integrated text assistant and is based on AlephAlpha’s Luminous language model, developed and hosted in Germany. The AI start-up developed the front-end and data processing especially for the City of Hamburg.

Around 100 city employees are testing out four functions: summary, research assistant, text generation and ‘expert mode’. The expert mode works similarly to OpenAI’s ChatGPT and users can freely interact with the system via self-defined prompts.

Theory to practice

Christian Pfromm, Chief Digital Officer for the City of Hamburg, said: “With LLMoin, we are moving from theory to practice in the key technology of Large Language Models and gaining valuable experience in the applicability of administration at an early stage.

“What we learn from this project will help us to align our future actions in this field with the actual needs of our employees.”

Hamburg is also carrying out wider activities to familiarise staff with generative AI tools. For example, in collaboration with the University of Hamburg, the city ran a ‘prompt-a-thon’ during an internal event. Employees were supported to solve problems with the help of ChatGPT.

The LLMoin project is backed by the InnoTecHH Fund, through which the Senate Chancellery supports authorities in Hamburg to develop and test ideas for the use of AI and other new technologies.

Cities and generative AI

Several cities around the world are experimenting with generative AI tools and creating policies for staff.

Boston Consulting Group recently forecast that the use of generative AI could create productivity gains of US$1.75 trillion per year in the public sector globally by 2033.

The analysts said the projection reflects productivity gains across all national, state and local governments and across all domains such as legislative, administrative, courts, healthcare, education, transportation and security.

The estimated benefits include US$519 billion a year across all governments in the US and US$128 billion in the UK. Germany and France could see gains of US$85 billion and US$83 billion respectively, and Canada is forecast to generate efficiencies worth US$73 billion per year.

The report identified the highest-potential use cases currently as summarising documents and meetings, reviewing and drafting procurement contracts, customer engagement with citizens and businesses, and developing and testing software code.

However, it said the conditions are not yet in place for government organisations to use generative AI at scale and highlighted risks related to accuracy, reliability and control, privacy and confidentiality, bias, and intellectual property ownership.