Beyond ChatGPT: What Else Can AI Become?

Most students encounter artificial intelligence through chatbots and image generators, but AI is quickly becoming much broader than that. In this session, we will explore some of the emerging possibilities for AI through live demonstrations and examples from current research.

We’ll look at questions such as: Can an AI model simulate an entire computer? Can a description written in English be turned directly into a working program? What can millions of real-world user-AI conversations teach us about how people actually use AI?

Along the way, we’ll also examine where today’s AI systems fail, why they can behave unpredictably, and what researchers are doing to make increasingly capable AI systems safer and more reliable. The goal is to give students a view of AI not just as a tool they use today, but as an active field of science whose future they could help shape.

Examples of research and demos I may draw from include NeuralOS (neural-os.com), where a generative model simulates an entire computer end to end, and ProgramAsWeights (programasweights.com), which turns natural-language specifications directly into neural programs.

Yuntian Deng

Yuntian Deng is an Assistant Professor of Computer Science at the University of Waterloo and an Associate at Harvard University’s School of Engineering and Applied Sciences. His research explores new ways of building and understanding artificial intelligence, including ProgramAsWeights (PAW), which turns natural-language descriptions into neural programs; NeuralOS, which uses generative models to simulate an entire computer; implicit reasoning in language models, which studies how models can reason through internal computation rather than expressing every step in human language; and large-scale studies of how people use AI. His WildChat research, based on millions of real-world conversations with AI systems, has been used by OpenAI and Anthropic for AI safety evaluation and featured in The Washington Post.

Bing Yan
Bing Yan is a PhD candidate in Computer Science at New York University and a Visiting Researcher at Meta FAIR. She previously earned a PhD in Chemistry from MIT, and her research brings artificial intelligence and chemistry together to help scientists organize scientific knowledge, model molecules and chemical reactions, understand reaction mechanisms, and design new experiments. She created AskChem, an open scientific knowledge platform that uses AI to organize and synthesize findings from the chemistry literature, and develops AI methods for understanding chemical reactions and guiding scientific discovery.