The purpose of this study was to scientifically explore the future of private generative LLMs through a user perspective, focused on the different aspects in which private LLMs could be more beneficial compared to public proprietary LLMs. During the past few years, the development and consequential usage of AI has increased exponentially, with public LLMs such as ChatGPT having millions of users. These public models however have continually come into disrepute for several reasons – the most common one being data privacy and how these AI tools handle user data. In recent years, however, the negative environmental impact of AI has gained more attraction, with researchers arguing it needs to be a major focus of the current AI conversation. Consequently, private LLMs offer a potential solution; being employed locally on a computer, they offer increased data privacy and control for the user, as well as possibilities for increased personalisation that can be catered more to individual needs. Finally, since they need not rely on energy- intensive data centres, private LLMs could prove to be a better alternative from a sustainability perspective.

The study was conducted with 15 participants aged between 20 and 30 years old, in a mixed-methods approach consisting of a questionnaire, an interaction study with a private LLM, and an interview. Thusly, the study examined people’s attitudes and stances on generative AI, with regards to the major themes of the study, the possible benefits and disadvantages of private LLMs for the future, as well as their awareness of data privacy and AI sustainability issues.

The results showed that people considered themselves mostly willing switch to a private LLM in the future, given the fact that they were easy to run and install, and that they gave adequate answers. Participants had quite low awareness of the environmental impact of AI, whereas more were aware of its handling of data. Nevertheless, everyone saw it as a strength of private LLMs that they could mitigate these issues. Personalisation proved to be more divisive, with certain participants liking the fact that one can train and change the characters, whereas some were uncomfortable with the anthropomorphic and more personalised traits. Furthermore, less than half of the participants had heard of private LLMs before the study, showing increased awareness is essential for both its existence and its potential benefits.