[rant] It Is an Existential Risk… Let’s Build It Anyway

You can build and deploy AI tools with a clearer understanding of their immediate engineering risks.

The article critiques the AI industry's approach to safety and risk, particularly regarding large language models (LLMs). The author argues that leading AI companies simultaneously proclaim their technology as an "existential risk" while aggressively racing to develop it, often with significant financial investment. This paradox is highlighted by the author's assertion that if a technology poses even a small probability of catastrophic harm (e.g., 1% chance of killing all humans), development should cease immediately due to the infinite expected value of harm. The author dismisses the idea that potential immense benefits justify accepting such risks. The piece suggests that many current AI risks are preventable engineering failures, citing the OpenAI-HF incident as an example of improper sandboxing and monitoring rather than an inherent existential threat. The author contends that AI companies may be leveraging the public's fear and the technology's mystique for PR purposes, advocating instead for responsible development with less fanfare and a greater emphasis on proactive safety measures from the outset.

The author suggests that the sensationalism surrounding AI's "existential risk" is partly a PR strategy, obscuring the fact that many current dangers are preventable engineering flaws. Companies are urged to implement safety measures like proper sandboxing and monitoring proactively, rather than as an afterthought or a public spectacle. This implies a shift towards more grounded, transparent safety practices in AI development, moving away from speculative doomsday scenarios towards addressing immediate, solvable technical issues.

This perspective challenges the industry's narrative by separating immediate, addressable risks from hypothetical future catastrophes. It suggests that if AI developers focused on robust engineering and responsible deployment now, they could mitigate many of the feared outcomes. The author's call for responsible development, even if it means slowing down, implies that the current breakneck pace might be contributing to the very risks being discussed, and a more deliberate approach is not only possible but necessary for sustainable progress.