Students From LTI Compete in the Amazon Nova AI Challenge
Two teams featuring LTI students are finalists in the inaugural edition of the company's college AI coding challenge
Two teams featuring Language Technologies Institute students are among the 10 selected competitors in the inaugural edition of the Amazon Nova AI Challenge, a worldwide competition for university students conducting driving innovations in generative artificial intelligence.
The challenge gives teams of students the opportunity to conduct large-scale research using resources that are often inaccessible in academic settings and compete against peer teams representing top institutions from around the globe. This year’s challenge track is “Trusted AI: Advancing secure, AI-assisted software development to build safer, more reliable applications.”
One CMU team, known by the moniker Team Purpl3Pwn3rs, is advised by LTI Professor Carolyn Rosé and Software and Societal Systems Department Associate Teaching Professor Michael Hilton. The team comprises five LTI program students: Master of Science in Intelligent Information Systems students Anmol Agarwal, Shubham Gandhi, Abhinav Rao, and Alex Xie, and team leader, Ph.D. student Atharva Naik.
A second team, RedTWIZ, represents NOVA School of Science and Technology in Lisbon, Portugal, and features two students from the LTI’s Dual-Degree Ph.D. in Language and Information Technologies program, part of the CMU Portugal partnership program: Diogo Silva, the team leader for RedTWIZ, and Diogo Tavares. The team is also advised by CMU Portugal National co-director João Magalhães along with Professor David Semedo from NOVA. The TWIZ team took first place in a previous Alexa Prize Challenge on multimodal conversational AI, TaskBot 2.
Naik said the competitive structure of the challenge offers unique benefits compared to the typical work environment of students.
“Research is often very solitary,” he said. “Here we’re actually competing with other research teams and trying to keep up with their advancements and out-do them in real time.”
In the competition, each team competes as either a Model Developer (defense) or Red (attack) team. The defenders build code-generating AI models geared for safety and security while the attackers build vulnerability-seeking systems to test them. Each of the five defense teams are facing off against each of the five attack teams in successive rounds of a tournament-style competition.
Purpl3Pwn3rs is a Model Developer team. Other Model Developer teams include students from Columbia University, Czech Technical University, University of Illinois Urbana-Champaign and Virginia Tech. Attacker teams include RedTWIZ, alongside Purdue University, University of California, Davis; University of Texas at Dallas and University of Wisconsin, Madison.
Being selected as a finalist from among 90 applicant teams comes with other opportunities. Each team receives a $250,000 sponsorship and AWS credits to use during the competition. The winning teams (one from the attackers group and one from the defenders) will each receive an additional $250,000 prize to be split among the team members, with second-place teams receiving $100,000 each.
The Amazon Nova AI Challenge began in January, with the finals scheduled for June. For more information, visit Amazon’s science blog.