Tentatively Trained Entropy Generators

In my intervention, I intend to explore the connections between information and entropy, and context – in post GPT-4 (large multimodal language model published by OpenAI in March 2023) commercialisation reality. We will begin with a “return to cybernetics”, its creator Norbert Wiener having warned, “Cybernetics is a two-edged sword, and sooner or later it will cut you deep”. Cybernetics is indeed worth revisiting, as it has revisited us. The contemporary context of cybernetic “image-objects” (such as the automatic factory described by Wiener) has little in common with the way of producing science or technology in the early post-World War II days. Nonetheless, the “image-objects” designed by cybernetic engineers have become a reality, yielding machines whose workings would cross all boundaries of cybernetic imagination: Amazon’s warehouses, autonomous vehicles, or the aforementioned GPT-4. With regard to the latter, I will argue that this “generative and tentatively trained transformer” is a primary-level trained generator of entropy – social and information-oriented – and of mental entropy defined by psychiatrist Antoni Kępiński. Yet in order to perceive these entropies and describe their generation mechanisms correctly, critical revision needs to be applied to the interpretation of entropy in information theory, and to the calculative description of the mind as a cerebral property involving information processing. I will attempt to showcase the outcomes of such theoretical interventions, as well as its potential practical use.