A new version of Italy’s AI response to ChatGPT, named Minerva7B, was unveiled Tuesday at Rome’s La Sapienza University.
Minerva7B, boasting seven billion parameters, more than double its predecessor, was developed by a team led by Roberto Navigli from Sapienza’s Natural Language Processing (NLP) group. Accessible to everyone at minerva-llm.org, the project is part of the Future Artificial Intelligence Research initiative.
Funded by the EU-backed National Recovery and Resilience Plan (NRRP), the project also collaborated with Cineca, a high-tech company that provided the Leonardo supercomputer.
“This is the first pre-trained Italian language model in the style of ChatGPT, designed for answering questions and following instructions,” said Navigli. “It is a model that is also trained to respond to questions and instructions, commands and it is particularly interesting because we have done important work on the security of the model, the so-called safety, that is, the ability of the model to avoid and respond to questions that could lead to the generation of unwanted or malicious content”.
Minerva7B is positioned as a key step in implementing Italy’s national AI strategy, blending advanced technology with robust security measures.
