RELATED TERMS: Promptography; Hallucination and Confabulation; World; World-Building
In recent years, Large Language Models (LLMs), such as ChatGPT, have become synonymous with the idea of artificial intelligence (AI). In order to train these chatbots, as Jack Apollo George (2024) notes, technology companies are employing human annotators or ‘data quality specialists’. The companies and the LLMs need examples of the kind of writing that the model will then emulate.
George points to two levels of irony in the situation as it currently stands. At the more immediate socio-economic level the irony is that the LLMs were developed in order to automate the task of writing. The better such models become at writing, the more rapidly the careers of human writers will decline and perish. Working as a digital annotator to improve the capacities of LLMs may be viewed as an exercise in self-destruction.
At the axiological level, LLMs as a socio-technical phenomena rewards and places a high value on language and writing. The irony here is that it simultaneously devalues writing as a human endeavour by viewing it as inferior and in need of improvement through ‘automating’ it.
In this latter case, it may be said to engage the logic of the supplement, as raised by Derrida (1976, 1981) and discussed by Nancy (2013) and Stiegler (2020). As Nancy explains, Derrida inscribes a twofold value into the logic of the supplement. Nancy and Stiegler talk of technology, but here we are displacing the notions of writing and technology onto that of ‘design’.
Thus, displacing Derrida, Nancy and Stiegler, the twofold value in question is that design supplants and supplements ‘nature’, here understood as that which is given, the existing, accepted, ‘fundamental’ state of affairs. Firstly, design comes to supplant or take the place of the given wherever what is given does not provide certain ends, for example a house or a bed. Secondly, design comes to supplement the given when it adds itself onto the ends and means of what is given.
Nancy contends that the supplement and its twofold movement, of displacing and adding to, falls under the category of technology, artifice, or art but to confirm what was said above, it is argued here that the supplement falls under the category of design.
Nancy further elaborates that two conditions are necessary for this supplementing to take place: the given state of affairs must evince some characteristic lacks, for example, while (undesigned) shelter may be offered the affordances of fully designed house are not; and it must be possible for designs to be grafted or collaged (‘engineered’) onto the given, using the prevailing (‘natural’) materials and the forces.
For Nancy, what is at stake in these processes is sense. He argues that whereas we were in the habit of relating sense to an ultimate purpose or final end, whether that be one of history, wisdom, or salvation, in a world pervaded by designs we are discovering that ends are proliferating at the same time as they are constantly transforming themselves into means. Paraphrasing Nancy, it can be argued that the ‘lesson’ of design is that through design nature itself, from which design is descended, shows that nature is by itself devoid of an end, without a ‘why’.
Language and Image
The emphasis on language is also evident in AI image generators or generative image AI platforms, such as, for example MidJourney. In this case, it is the relationship between word and image that is at stake. In ekphrasis, a vivid description or poem is written in response to a visual work of art. In reverse ekphrasis, in which AI image generators engage, visual representations are created in response to the written word, as discussed in the post Promptography.
Large Language Models and (Superhuman) Intelligence
Yann LeCun, Turing Award winner and one of the pioneers of artificial intelligence, argues that LLMs are useful but crucially limited and constrained by language. Achieving human-level intelligence requires an understanding of how the physical world works in addition to the capabilities of language. To address this challenge, LeCun has developed an architecture called V-JEPA, a world model that aims to understand the physical world by learning from videos and spatial data, in addition to language. LeCun labels this kind of intelligence Advanced Machine Intelligence (AMI), a kind of intelligence that is able to plan, reason and have persistent memory, enabling it to rely on past experience and evaluations to guide its decisions, characterised as a kind of ’emotional’ response (Heikkila, 2026).
Given this stance, LeCun concludes that LLMs, if taken by themselves, are dead ends when it comes to developing ‘superintelligence’ or ‘superhuman’ intelligence.
References
Derrida, J. (1976). Of grammatology. Baltimore, MD: Johns Hopkins University Press.
Derrida, J. (1981) Dissemination. Translated by B. Johnson. London, UK: Athlone Press.
George, J. A. (2024), ‘If journalism is going up in smoke, I might as well get high off the fumes’: confessions of a chatbot helper. The Observer, 7 September. Available at https://www.theguardian.com/technology/article/2024/sep/07/if-journalism-is-going-up-in-smoke-i-might-as-well-get-high-off-the-fumes-confessions-of-a-chatbot-helper [Accessed 15 September 2024]
Heikkila, M. (2026). “Intelligence really is about learning”. Financial Times, 3 January, Life & Arts p.3
Nancy, J.-L. (2013) Of Struction, Parrhesia. Translated by T. Holloway and F. Méchain, (17), pp. 1–10.
Stiegler, B. (2020) ‘Elements for a general organology’, Derrida Today, 13(1), pp. 72–94. doi: 10.3366/DRT.2020.0220.
Incomplete ... 