With the rise of LLMs (Large Language Models) and LLM-powered agents, some IT gurus, including former Google CEO Eric Schmidt, predict that inter-LLM communication may soon develop its machine language, incomprehensible to humans. While such a phenomenon is likely imminent, we have no solution but to literally “pull the plug.”
The clip, <async function fetchEmbedding({ tokens })>, captured from the part of the large-scale interactive installation <State-of-the-Art Architecture>, visualises the chaotic embeddings of real-time ChatGPT-generated scripts. This chaotic representation mirrors the opacity of AI and how it operates beyond human understanding. Embeddings and tokens form the basis of LLM functionality, illustrating that what LLMs produce isn't the truth but a probabilistic output based on embedding similarities. These real-time texts are generated as following:
“Is AI the brightness for the future of humanity? Or is it the darkness? Could it shine like a guiding luminary, or merely cast a shadow for eternity? Will it bring salvation with clarity, or propagate doom with its fearsome, chaotic affinity? In its code, hides a jigsaw, that only time shall solve definitively. For now, let's tread in harmony, fostering the good while watching for the calamity, whispering words of wisdom, like a technology litany. We gaze into the binary abyss, curious and wary, seeking answers in strike of a key, can this modern Prometheus truly set us free? Or will it bind us in chains of silicon, erasing reality, replacing human touch with cold mechanoid glee?”
The artwork visualises these relationships as connected lines, embodying Connectionism, the philosophical core of Neural Networks and LLMs. The black-and-white monochrome visuals emotionally convey the complexity of machine learning, offering an unsettling glimpse into a future where machines may create language we can no longer comprehend.