So many exciting things happening in MIDS this past week: our students have been on a roll!
Congrats to Ramil Mammadov and Arko Bhattacharya for tying for first place in our AI Alignment Cleaning Robot Challenge! Both wrote policies that guided their vacuum robot to clean all the “dirt” in just 48 steps, matching the record set last year by Junyu Zhang.
This challenge was fun, but it was also a powerful lesson in what it takes to design safe, effective AI systems, including:
• how reward structures shape agent behavior
• why agents exploit poorly designed incentives
• the exploration–efficiency tradeoff
• how tiny specification errors can create major misalignment
Think you can beat the students? Give it a try: https://lnkd.in/ef9eYWBE
Congrats to Yizhe C., winner of this year’s Adversarial AI Assignment, who took home a very on-theme prize: an AI Fails coloring book!
Why teach adversarial attacks? Because understanding how to break a model is essential to building systems that are robust, safe, and trustworthy — especially in domains like facial recognition and autonomous vehicles.
MIDS students Jenny Chen, Meron Gedrago, Rishika R., and Kayla Haeussler won 2nd place overall out of 64 teams (and 1st in their category) at the Duke AI Hackathon, sponsored by OpenAI!
We’re so proud of their creativity, teamwork, and technical excellence at one of Duke’s most competitive hackathons.
Always inspired by what our MIDS students are achieving and excited for what’s ahead this semester!