TCube presented to SISSA mathLab applied mathematics group
The virtual platform of Trieste Lab Village, based on generative AI, meets its first users
Trieste, 19th July 2024
On 12th July, the members of the SISSA mathLab research group, focused on mathematical modeling and scientific computing for practical applications, participated in the session named TCube – Empowering Tech Transfer with GenAI @SISSA. This was a preview presentation of TCube (Trieste Tech Transfer), a digital platform based on generative artificial intelligence that will serve as the virtual component of Trieste Lab Village. This marks the first step towards engaging early adopters of the platform, which aims to unlock Trieste’s full potential in the field of technology transfer: “We need to build a community of highly engaged users who can use the platform on a daily basis and promote it,” says Ottavio Crivaro, manager of the development of TCube.
Following the opening by Prof. Gianluigi Rozza, Spoke 9 coordinator, there was a speech by Paolo Ferrandi, Head of Data Valorization at RES Group, devoted to developing the platform. As Ferrandi explained, TCube has a dual purpose: abstraction and translation. Starting from a specific numerical method, TCube will expand its application scope, both within the sector where it is already employed and in other industrial sectors (abstraction). Then, the numerical method will be made accessible to the industrial world, aligning it with its needs (translation). All this will be possible thanks to TCube’s generative AI system, based on Large Language Models (LLMs), algorithms capable of generalizing and translating from previous knowledge.
To ensure TCube becomes a functional and effective tool, its developers will need to tackle some technical challenges: firstly, LLMs lack logical principles or the concept of truth—leading to incorrect interpretations when provided with common knowledge expressed in common language. Additionally, these algorithms are extremely efficient in synthesizing and enriching text within a specific domain but are not yet capable of abstracting its content. Lastly, there is the risk that allowing completely free-flowing conversation could cause the AI to significantly diverge from the user’s desired answer.
TCube will be officially launched at the end of October, but it will be available in advance to the members of the SISSA mathLab group at the end of July.
