-
@ halalmoney
2025-05-14 10:05:29
"…machine learning programs are speculative and analytic, based on brute force approaches to optimization and repetition, and they do not presume any relationship between speaker and audience (see M, pp. 19, 35). The “learning” that underlies these systems is not one based on educating or dialectical persuasion but rather instruction and demonstration. Their methods are admittedly crude—an inefficient and inelegant “training” based on quantity rather than quality. Neural nets entail seemingly repetitive calculations to minimize loss functions—they epitomize the triumph of numerical methods and brute force repetition over twentieth-century expert and law-based systems. Their results point to a deeper more primary and unknowable latent space, which seems a perverse mirror of language and nature, in which there are, as Saussure once argued, no positive terms—hence the multiplicity of tokens that make “understanding” Large Language Models (LLMs) such a challenge. Everything is defined in negation and in them, repetition seems to run wild—unhinged from meaning and context. Supervised learning—in which a system “learns” the correct response—reduces learning to guiding systems to repeat the right answer, so that it can classify what it has not been trained on correctly. Even attempts to make these systems more efficient such as reinforcement learning, which has been deployed by systems such as Deep Seek take inspiration from behaviorist experiments based on reward-punishment that were nothing if not cruel. The cycling of reward and punishment to help humans and animals learn the right answer seems not only crude but also a travesty to learning itself."
Our Cruel Crude Techne?
Wendy Hui Kyong Chun