Questions 1–5.

Questions 6–10. Select the TWO correct answer choices.

Questions 11–15 are based on the passage below.

Large language models, trained to predict the next word in a sequence of text based on statistical patterns in enormous datasets, have proven capable of tasks their designers did not explicitly anticipate, including basic arithmetic, translation between languages never paired in training, and rudimentary logical inference. This phenomenon, sometimes labeled "emergent capability," has generated substantial disagreement among researchers about its proper interpretation. One camp views emergent capabilities as evidence that sufficient scale in model size and training data produces qualitatively new abilities, a genuine emergence analogous to how complex behavior arises from simple rules in other systems. A competing camp argues that apparent emergence is largely a measurement artifact: capabilities that seem to appear suddenly at a certain model size may in fact be improving gradually all along, with the "sudden" jump reflecting only the coarseness of the metrics used to evaluate performance rather than any genuine discontinuity in the underlying ability. Under this second view, a smoother evaluation metric would reveal steady, unsurprising improvement rather than a genuine qualitative leap. Resolving this disagreement matters beyond mere scientific curiosity: if capabilities can emerge unpredictably at scale, safety researchers argue, future systems might acquire concerning abilities without warning, whereas if improvement is genuinely gradual and predictable, such abilities could in principle be anticipated and prepared for well in advance.

Questions 16–20 are based on the passage below.

The history of automation anxiety long predates artificial intelligence: nineteenth-century textile workers, twentieth-century assembly-line laborers, and now knowledge workers have each, in turn, confronted fears that machines would render their labor obsolete. Economic historians studying these earlier waves note a recurring, if imperfect, pattern: automation has historically destroyed specific jobs while simultaneously creating new categories of employment that were impossible to anticipate in advance, from textile-machine repair technicians to entire industries built around the automobile. Critics of applying this historical pattern to the current wave of AI-driven automation argue that previous transitions, however painful for the individuals displaced, typically unfolded over one or two generations, allowing labor markets and educational systems time to adjust; the pace of AI adoption across many industries simultaneously, they contend, may compress this adjustment period so severely that the historical precedent offers little comfort. Proponents of the historical-parallel view counter that predictions of technological unemployment have recurred at every prior wave of automation and have so far proven premature each time, suggesting a persistent tendency to underestimate the economy's capacity to generate new forms of work even when the mechanism by which it will do so cannot yet be specified.