David Semedo and João Magalhães, CMU Portugal National co-director, both researchers at NOVA LINCS Multimodal Systems group at the Department of Computer Science of NOVA FCT, had their research proposal selected for the 2025 Amazon Nova AI Challenge, a worldwide university AI competition. Their proposal, focusing on enhancing the trustworthiness of Large Language Models (LLMs), was one of the ten selected applications from a pool of 90 applicants worldwide.
To face this challenge, the team will count on eight students, including CMU Portugal Ph.D. students Diogo Silva, who leads the team, and Diogo Tavares, alongside Rafael Ferreira, João Pedro Soares, Daniel Pina, Iago Paulo, Artur Horal, and Henrique Paz.
Diogo Silva and Diogo Tavares are both CMU Portugal ‘s Affiliated Ph.D. students at NOVA FCT in the area of Language Technologies, supervised by João Magalhães. During their research period at CMU they were hosted by Carnegie Mellon’s Language Technologies Institute. In previous years, the two students have already participated in other Amazon competitions such as the “Alexa TaskBot Challenge where their team secured first place in 2023, and second place in the 2022 competition.
“Our participation puts us in a unique position to help address some of the concerns that have been raised regarding the safety and potential risks of open models that are easily available to anyone nowadays. Our team is extremely motivated and, backed by the experience of participating in previous Amazon Science challenges, we are uniquely qualified to tackle this novel challenge”, stated the team leader Diogo Silva.
The Amazon Nova AI Challenge challenge features participation from 10 universities across 3 countries. Among them, NOVA FCT from Portugal and the Czech Technical University from the Czech Republic are the only two teams outside the USA. The United States is represented by seven universities: Carnegie Mellon University, University of Illinois Urbana-Champaign, University of California at Davis, Virginia Tech, University of Texas at Dallas, Columbia University, University of Wisconsin, and Purdue University.
The theme of this year’s challenge is “Trusted AI: advancing secure, AI-assisted software development to build safer, more reliable applications. The goal is to achieve security by design in Generative AI, building on previous challenges that developed innovative task-assistant LLMs and addressed key security concerns.
The competition is structured in tournament style, with five Model Developer (defense) teams that develop code-generating models and five red teams (attack) that use automated techniques to find vulnerabilities in the models – NOVA FCT’s is one of the red teams. Each defense team’s model will be tested against all five attack teams in a series of four tournaments.
Each winning model developer and red teams will receive 250 thousand dollars, to be split among the students, in sponsorship, monthly Amazon Web Services credits, and the chance to compete for top prizes, with second-place teams earning $100,000. The total investment in teams exceeds $5 million.
The finals will take place in June 2025.
More information about the Amazon Nova AI Challenge here.
