The aim of this internship is to investigate ways to fine-tune LLMs to workplace-specific language and optimize the output to situation-specific context. For example, when a patient converses with a doctor, it’s crucial that the conversation is accurately summarized in a way that can be easily comprehended by the patient, yet also be thoroughly documented by the doctor. The output of the LLMs can then be used for further analysis to understand the whole dataset and find patterns. We also want to test the different existing LLM models (LLaMa, Bard and GPT-3/4) before and after fine-tuning. Our goal is to determine which LLM model is best fine-tuned with jargon to yield benefits in day-to-day situations. During your internship, you will have access to different types of data, including:
- Legal texts
- Transcripts of meetings
- Conversations with a customer service agent
- Medical texts
UP FOR A CHALLENGE?