Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : May 29, 2024 - June 01, 2024
In recent years, the global food crisis has intensified, posing a challenge to secure sustainable food supplies. Factors such as population growth, climate change, and labor shortages in primary industries threaten the stability of food production. In response, insectbased food has gained prominence as an efficient and sustainable food source, offering benefits such as resource efficiency, environmental friendliness, and high nutritional value. This study explores the capability of a largescale language model (LLM), like ChatGPT, to learn safe cooking methods for insectbased food. Despite ChatGPT's general ethical constraints, we utilize the Finetuning API from OpenAI to train the model using a dialogbased dataset. Information about insects and poisons is collected from Wikipedia for training purposes. Evaluation of the model's output is performed using OpenAI's evals, specifically examining the inclusion of appropriate cooking methods. The study assesses the model's performance over 3, 10, and 30 epochs for 50 different types of insects.
