Sidony O’Neal

This video brings together two excerpted texts, a sculpture, and a performer. First the texts. One is my own translated excerpts from the short fiction work Funes, el memorioso by Jorge Luis Borges (1942); the second is snippets from the machine learning research paper “Attention Is All You Need” (2017). . Published 75 years apart, both texts allegorise and demonstrate techniques in memory management, encryption, and resource optimisation. Both texts reflect on the accumulation of hidden steps required to reason acceptably in time. Both texts review some stakes for optimising computational abstraction.

Published in 1942, Funes, el memorioso functions as a eulogy for a 19-year-old Uruguayan priest who experiences full presence and memory without management or compression—that is RAW memory—and thus remains secreted away from the overwhelming redundancy of the world. Incapable (or uninterested) in performing the fundamental abstractions of human consciousness expected by interlocutors, Funes declares “My memory, sir, is a landfill.” The cuento is of course expanded through Borges’ nuanced treatment of the Uruguayan and Argentinian social and political contexts from which it was conceived. Ficciones is Borges’s collection published in 1944 containing Funes, and it remains a bare entryway to the study of 20th century literatures as well as the applied philosophy of mathematics. Thoroughly addressing and innovating through genre, though not always gender, the Borgesian register rings dutiful with the sort of IYKYK refs, algorithmically strolled corpus, and emergency brevity that would not be out of place communicating on today’s internet.

“Attention is all you need” is a machine translation research paper published by a team of researchers—Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin—many of whom worked at Google—in 2017. It contains a succinct and playful proposal for a new architecture in machine translation that eschews the redundancy and inefficiencies from previous approaches. The paper quickly found application across several fields and notably formed the foundation for the current cultural shift (and mass investment) in the proprietary advancement of contemporary Natural Language Processing (NLP) products—including neural networks, machine learning modules, and their related areas of research. The paper motivates and demonstrates a layered “transformer”/translation structure that uses a multi-headed form of “self-attention” to exercise priority in learning. Only eight years after being published, “Attention is all you need” was recently counted among the ten most-cited scientific works of the 21st century. Its methods remain integral to contemporary large language models that form the basis for products such as ChatGPT and BERT.

Together, both texts provide a glimpse into some of the research space and themes present in my 2023 exhibition ,The Pudding Butcher, which brought together works and research undertaken during my time as a Hodder Fellow at Princeton University (2022-2023). These two texts take up the stuff of machine-like human and human-like machine relations. They thus invoke ability/disability and neurodivergence as well as computation and proprioception–topics complicating the desire for innovative machine technologies that can truly engage the plurality of computing bodies navigating an exceedingly detailed world. Putting these texts in touch again allows me to revisit and extend my investigations in that body of work, as well as further my engagement with a sculpture that I made nearly a year after completing the fellowship.

To make a secret, attention is all you need is a short film that concatenates and softly misreads both texts (Funes and “Attention”) by making them narrate the world of a sculpture I showed in 2024: A secret is not a substance (Hyperbolic Settee). The Hyperbolic Settee appears in the video, unfixed, in relation with dancer, performer, choreographer, and curator keyon gaskin, who embodies and exploits the sculpture’s layered formalisms. I am interested in the increased tendency toward self-surveillance/auto-attention that engagement with NLP products brings. An unattainable desire for lossless computational memory is in tension with anxieties about reconstructive (sur)veillance, search history, and self-snitching, which often have legal and material impacts. Both performer and sculpture treat the attention function as at once modular and redundant in the given environment/field:

Here, the conditions for the use of encryption technology are reduced to the basic human desire to keep a secret. The desire to maintain a secret first requires the production of said secret. We set out to unpack what it might mean to make a secret in an unfamiliar environment.

— Sidony O’Neal