Local language models writing haikus live in a gallery.
Three NVIDIA Jetson Orin Nanos running a local language model and RAG system generated haikus in response to user input. RAG (retrieval-augmented generation) was used to guard the output from potential malicious input - rather than using user input directly, the system retrieved relevant quotes from a large curated set of quotes and used those to generate the haiku. The result was similar enough to the user's response to be satisfying, different enough to be thought-provoking, and guaranteed safe from malicious prompt injection.






