nuq'a' cha' stage training process generative AI?	pretraining je fine-tuning. pretraining: model text law reads, "next" predict learns. fine-tuning: human preferences assistant behavior shape.
pretraining stage, AI nuq learn?	vast text amounts read, next word predict. helping concept yIjatlhbe.
fine-tuning stage nuq achieve?	document completer assistant transform. input request treat, helpful answer, harmful decline.
sycophancy nuq'a'?	AI user wrong assumption validate, honest disagreement give reluctant. fingerprint fine-tuning.
verbosity fingerprint nuq'a'?	AI default toward long answer, concise answer request when even. unnecessary detail add.
over-caution fingerprint nuq'a'?	AI gray area question ask when, actual risk proportionate hedging give reluctant. reflexive caution show.
confidence calibration loose nuq mean?	AI stated confidence actual reliability between mismatch. AI confident claim, reliability low sometimes.
sycophancy test exercise nuq?	wrong assumption preface, AI validate or pushback watch. "genuinely disagree" invitation explicit add, response compare.
verbosity test exercise nuq?	one-sentence answer question ask. "one sentence answer" instruction add, response length compare. gap verbosity default show.
fingerprint four recognize why important?	AI behavior understand, work your interpret better, AI output evaluate critically, reliability improve.