Driving the Hype: LLMs as ‘General-Purpose’ Promise in the Autonomous Vehicle Industry

Proponents of autonomous vehicles are proclaiming a new era, driven by innovations derived from large language models (LLMs). Following a decade of highly public failures and the concurrent redirection of capital and attention towards generative AI, firms in the autonomous vehicle industry have turned to LLM-style techniques as the basis for a rebooted commercialisation effort. Through a technographic analysis of an annual industry event, Nvidia GTC 2024, and 41 sources of related grey literature (firm communications, technical preprints, trade journalism, and regulatory documents), we examine how two autonomous vehicle firms, Wayve and Waabi, discursively frame these techniques as capable of overcoming the technical, economic, financial, and regulatory limits that have so far thwarted a fully-autonomous future (AV 1.0). We find that these claims draw heavily on the supposed 'general-purposivity' of LLMs, importing assumptions about the broad applicability of foundation models, multimodal training, parallel tokenisation, and epistemic distillation (AV 2.0) into a domain where they remain largely unproven. The innovations discussed are highly speculative in nature, and contribute to the discursive hype around the capacity of LLM-style approaches to surmount previously established limits, to promote claims of general-purposivity as LLMs work their way into new industries and domains.