Published: January 2025
Generative AI has the potential to transform the ways we live and work. It offers opportunities to increase productivity and improve decision-making; however, there are also associated risks and limitations users need to be aware of.
Between April and September 2024, AMRC hosted Julia Machalska, a civil service fast streamer to investigate the topic. As part of her time with us she developed the following set of briefing notes to serve as an introduction for charities and individuals looking to start familiarising themselves with this topic.
Please note that due to the rapid pace of technological advancements and ongoing developments in the field, these briefing notes reflect the current state of generative AI as of October 2024 and represent a point in time.
Generative AI is a type of artificial intelligence (AI) that can produce novel content such as text, images, audio or code. Currently popular generative AI tools such as ChatGPT, Google Gemini or Microsoft 365 Copilot use algorithms called Large Language Models (LLMs) that are capable of processing natural language to understand user inputs and generate coherent responses.
Simply put, generative AI tools respond to user inputs by predicting probable patterns and outcomes based on the data set they were trained on and then generating a response that they decide is the most likely. This decision-making process is extremely complex and is determined by millions of parameters that influence the output.
The input shared with a generative AI tool instructing it to carry out a specific task is called a prompt. An example of a text prompt is “Suggest a structure and sub-headings for a report discussing how medical research charities in the UK demonstrate research impact”. Generative AI tools rely on the information included in the prompt to generate a response, so including additional details will result in a more tailored output. Users can also ask the tool to alter the generated output with follow-up prompts, such as “Could you refine the proposed structure to make the report focus on small charities?” or by providing additional context such as the intended audience for the tool to consider.
Generative AI tools can be separated into different categories based on how they are accessed by the user and what data they have been trained on and have access to. While off-the-shelf tools are available from third-party developers, it is also possible for organisations to commission or build bespoke models for internal use. Types of tools include: