Uploaded November 2026 | Updated November 2026, 3 hours ago
Three Stanford scholars share what they are working on in three-minute lightning talks.
00:29 *Expert-Guided Bayesian Optimization for Sustainable Protein Design*
Anna Thomas presents a method in which an expert, such as a human or an LLM, guides Bayesian optimization to design plant-based protein products. Applied to two plant-based dairy products, it improved utility by 29% and 26% in 10 iterations each.
03:36 *Fantastic Proteins and Where to Find Them*
Gowri Nayar presents a model that predicts where proteins are located in the cell, including proteins that imaging has not yet measured. The predictions extend 223 cellular systems with more than 3,000 previously unmapped proteins.
07:21 *LLMs Outperform Humans at Personal Attribute Inference: Implications for Proxy Discrimination Risk and Equal Protection Law*
Victor Wu benchmarked 34 LLMs against a nationally representative sample of 1,018 U.S. adults and found that LLMs systematically outperform humans at inferring personal attributes from text. He argues that current U.S. equal protection doctrine is ill-equipped to address the resulting proxy discrimination risks.
_Confronting Our AI Future: Hope, Fear, and the Choices Ahead_ is a Stanford HAI conference examining the anxiety that AI's rapid acceleration has raised for students, workers, and leaders, and the choices that will shape what comes next. It took place on October 28, 2026, on the Stanford campus.
Three Stanford scholars share what they are working on in three-minute lightning talks.
00:29 *Expert-Guided Bayesian Optimization for Sustainable Protein Design*
Anna Thomas presents a method in which an expert, such as a human or an LLM, guides Bayesian optimization to design plant-based protein products. Applied to two plant-based dairy products, it improved utility by 29% and 26% in 10 iterations each.
03:36 *Fantastic Proteins and Where to Find Them*
Gowri Nayar presents a model that predicts where proteins are located in the cell, including proteins that imaging has not yet measured. The predictions extend 223 cellular systems with more than 3,000 previously unmapped proteins.
07:21 *LLMs Outperform Humans at Personal Attribute Inference: Implications for Proxy Discrimination Risk and Equal Protection Law*
Victor Wu benchmarked 34 LLMs against a nationally representative sample of 1,018 U.S. adults and found that LLMs systematically outperform humans at inferring personal attributes from text. He argues that current U.S. equal protection doctrine is ill-equipped to address the resulting proxy discrimination risks.
_Confronting Our AI Future: Hope, Fear, and the Choices Ahead_ is a Stanford HAI conference examining the anxiety that AI's rapid acceleration has raised for students, workers, and leaders, and the choices that will shape what comes next. It took place on October 28, 2026, on the Stanford campus.










