​​In archival research, generation loss refers to the gradual degradation of quality that happens in the physical reproduction of analog media, leaving out digital reproduction, where quality stays consistent.

‍

Yet, digital reproduction in generative AI follows a different logic. Copies often involve the re-introduction of a prompt rather than the preservation of a generated result, and in some cases, they fail produce an output altogether. I generated AI outputs from complex, unconventional prompts—such as electric hums, woodwind and string multiphonics, and extended techniques of the voice. When faced with prompts outside its training, commercial AI often responds with an empty audio file. Yet, by compressing the audio of these “empty” files, I was able to reveal a residual signal—a dense, gritty noise that prevailed from the failure of generation. In this piece, I explore the sonic consequences of generation loss both from the degradation caused by mechanical reproduction and the "unsuccessful" outputs of generative AI.