Table C.2: Summary of ground-truth clusters used in the experiments

One day last spring, in a high school classroom in Texas, students were arguing about who to kill off first. It was a thought experiment with a sci-fi premise: A global zombie outbreak has decimated major cities. One hundred frozen embryos meant to reboot humanity are safe in a bomb shelter, but the intended adult caretakers never made it. Instead, 12 random civilians stumbled in. The students had to decide who would die and who would live to raise the future of the human race.

If you're a parent, an educator, or just someone who's been to school, you've probably developed an opinion about generative AI in classrooms. You might fear the demise of the five-paragraph essay, the ever-increasing ease of cheating, or, worse, the end of critical thinking altogether. But don't worry: The anxiety surrounding large language models in schools is anything but unprecedented. In 1975, teachers fretted that handheld calculators would undermine students' capacity to "handle basic skills like reading, writing, and arithmetic," according to a report in The New York Times. Others, though, believed calculators could "free students to concentrate on basic principles."

However, there are still settings where their zero-shot performance is far from optimal.

Emerging ethical approaches have attempted to filter pretraining material, but such approaches have been ad hoc and failed to take context into account. We offer an approach to filtering grounded in law, which has directly addressed the tradeoffs in filtering material.

We introduce a simple but general online learning framework in which a learner plays against an adversary in a vector-valued game that changes every round. Even though the learner's objective is not convex-concave (and so the minimax theorem does not apply), we give a simple algorithm that can compete with the setting in which the adversary must announce their action first, with optimally diminishing regret.

(since we pick the log base to be 2). We now prove the converse statements. Individual fairness aims to ensure that "similar individuals are treated similarly." Predictive multiplicity allows different predictions from competing classifiers for the samples. Notably, neural networks with very narrows or wide layers have better reproducibility in their decision regions. The fact that multiple classifiers may yield distinct predictions to a target a sample while having statistically identical average loss performance can also cause security issues in machine learning.