In the VLAIO-TETRA project AI CARES, Artevelde University of Applied Sciences, Thomas More University of Applied Sciences and Ghent University are developing chatbot prototypes trained on specific knowledge sources from wellbeing and mental health. In several use cases, we test different techniques and explore, together with frontline organisations, how AI chatbots can conduct better conversations about psychosocial topics. In doing so, we experiment with system prompting, RAG and LoRA fine-tuning. With system prompting, we give the chatbot instructions in advance about its role, tone, boundaries and way of working. With RAG, we allow the chatbot to retrieve relevant information from selected knowledge sources, so that its answer is better aligned with reliable content. With LoRA fine-tuning, we make targeted adjustments to the underlying language model using example data, without retraining the entire model.
Through the WatWat chatbot, we investigate how a chatbot can tailor information more effectively to the question and situation of the person seeking information, without compromising the quality of the content. The Triple chatbot demonstrates the effect of imposing ethical boundaries on sensitive topics. The Training Chatbot developed with SAM vzw makes it possible to practise chat counselling conversations through role play with a chatbot and receive personalised feedback. The CLB Finetuned chatbot allows us to compare whether adapting the underlying language model itself — by further training it on the basis of 14,000 example conversations — also leads to better chat conversations.
Try it out for yourself and draw your own conclusions... Feel free to let us know what you think.
Tim Vanhove (tim.vanhove@arteveldehs.be) George Caique Gouveia Barbosa (GeorgeCaique.GouveiaBarbosa@UGent.be)