An analysis of land degradation hazard applying a back propagational neural network using Geographic Information System (GIS) map datasets was conducted in 1995. The entire land area of 903 km2 was inferred for the land degradation, resulting precisions of estimation at 86.4% and 79.4% in degradation degree and extent respectively, with an error margin of 0.25 against the supervisors. A Large Language Model (LLM) fine-tuning was conducted to create catchphrases for university festivals, applying prompt engineering in 2023. The comparison of LLM models with/without fine-tuning was conducted through Google Colaboratory exercises. Through this, students experienced that the instructional texts should be specific and found that generative AI models were not able to work without intensive human training. A clustering study of Sue wares of sixth centuries produced in Japan was conducted. The Vector Quantized Variational Auto Encoder (VQ-VAE) with 16,384 latent variable dimensions, a model with a learning rate of 0.001 and for 300 epochs, made three distinct groups. One cluster consists solely of type II-5, another cluster contains a mixture of types II-3, II-4, and II-5, while the third cluster is centered around types II-1 and II-2, with types II-3 or II-4 mixed in toward the edges.