Packet M: Tossup 11

Description acceptable. Over 100 implementations of a strategy for this task are surveyed in a 2024 paper by Gao et al. that categorizes them as Naive, Advanced, or Modular. Novel embeddings of query-specific documents are indexed and used in the “retrieval-augmented” form of this task. (10[4])This (10[1])task is performed (10[1])by applying Brownian noise to points in a latent (10[1])space (10[1])and then decoding (10[1])in diffusion models. (-5[1])A game-theoretic (10[1]-5[1])agent improves at this task while competing against a discriminator in a type of “adversarial” neural (10[1]-5[1])network. (10[2]-5[2])This task (10[1])names a framework consisting of unsupervised “pre-training” followed (-5[1])by supervised (-5[1])“fine-tuning.” (-5[1])Nonsensical outputs during this (10[1])task are (10[1])called “hallucinations.” (10[4]-5[2])For 10 points, what (10[1])task (10[1])names a form of AI commonly (-5[1])used to produce text and images (10[2])that is (10[1])the “G” (10[2])in GPT? (10[1])■END■ (10[10])

Back to tossups

Buzzes

Summary