Yeh Tinghao From taipei
( 葉廷皓 )
Yeh is a new media artist who works in the field of creative coding, immersive visuals, experimental sound and live performance . His works mostly focus on the relationship of sound and visual, heading to integrate prefabricated and instantaneous events, starting from glitch aesthetics, and attempting to find parodies and dialectics belonging to the new media age. In addition to his solo works, he also participates in dance and theater productions. Since 2013, he has formed the Audio-Visual duo HH with YAO Chung-Han. He is currently the head of the experimental audio-visual promotion group “Fluid Noise”, selected as Artist-in-Residence at National Taichung Theater 2023-24 and concurrently teaches in relevant departments of many universities.
新媒體藝術家，擅長創意程式設計、沉浸式視覺、實驗聲響與現場演出。作品多聚焦於聲音與影像之間的關聯，融合預製與即時發生的事件，以錯誤美學為出發，企圖尋找屬於新媒體時代的諧擬與辯證。除了個人作品之外，也參與舞蹈與劇場製作。2013起與姚仲涵組成Audio-Visual 團體HH。目前為實驗音像推廣團體「噪流」現任負責人，並獲選臺中國家歌劇院2023-24駐館藝術家，目前在實踐大學媒體傳達設計學系任教。
Alchemy of Non-Human Transmutation
Large language models like ChatGPT, or text-to-image models like Stable Diffusion, are developed through a process of training on vast amounts of data, compressing the original data into highly versatile models. Once these models are established, users can refine the output through prompts, reduce noise, and restore a sense of reality, ultimately obtaining a result that closely aligns with their imagination.
We can consider the operation of these models as a form of collective unconsciousness among machines. When users pose questions to the machine, the machine’s responses sometimes exhibit AI hallucinations. These hallucinations stem from the model’s unconscious processes; they are not the machine’s conscious thoughts, but rather approximations generated by comparing and denoising information within the model based on the prompts. Whether or not these models produce outputs that are close to reality, they ultimately transcend reality, approaching the manifested objects in the human consciousness. This process is akin to an alchemist iterating materials to transmute them into gold.
But can alchemy truly transmute any material into gold?
大型語言模型如ChatGPT，或是文本轉換圖片模型如Stable Diffusion，都是透過訓練資料的過程，高度壓縮原始資料，成為幾乎萬用的模型。成為模型之後，使用者透過提示詞提煉結果、降低雜訊、還原真實，最終將得到一個趨近於使用者想像的結果。
我們不妨將這類模型的運作過程視為機器之間的集體潛意識。當使用者向機器提問，機器的回饋出現AI幻覺，這些幻覺源自於模型的潛意識；這些幻覺並非機器意識本身，而是機器根據提示詞，從模型內比對、降噪，最終生成的近似值。無論這個模型能否產生接近真實的圖文，終將超越真實，逼近人類意識世界中的具現物。而這個過程，就好比術士透過鍊金術，將其他原料迭代成黃金。
但鍊金術真的能將任何原料，幻化成黃金嗎？
Medium
Video
