P(doom)

The availability of powerful AI models is becoming increasingly expensive for many industries and researchers.

Armin Ronacher, in a blog post, questions the prevailing narrative and proposed solutions surrounding the "AI doom" scenario, particularly referencing concerns raised by Dario Amodei, Sam Altman, and Elon Musk. While acknowledging the validity of observations regarding AI's current impact, Ronacher expresses opposition to the idea that AI development needs "pacing," especially when initiated by a few large, closed-weight labs like OpenAI and Anthropic. He argues that these companies, which benefited from public data, are now straining public resources and creating an economic imbalance. Ronacher points to issues like botnets and the poisoning of platforms like RubyGems as current problems, but suggests that the primary worry should be the detrimental effect of these closed, subsidized AI models on individuals outside of their development. He contrasts this with the inherent pacing mechanism of open-weight models, advocating for broader accessibility to mitigate current issues and prevent geopolitical imbalance.

Ronacher critiques the "pacing" discussion as primarily benefiting a duopoly of closed-weight labs, OpenAI and Anthropic, which have leveraged public data and are now straining resources. He posits that true pacing arises organically from open-weight models, which democratize access and innovation, thereby leveling the playing field. The current system, he argues, allows these labs to operate at a massive loss, distorting markets and creating an economic feedback loop where society indirectly subsidizes development only to buy back benefits from a few powerful entities.

He asserts that current AI-related issues, such as the poisoning of code repositories and the strain on public infrastructure, stem from these closed models, not open ones. Ronacher suggests that regulatory failures in Europe and the US exacerbate these problems, leading to a chaotic market resembling a drug economy. He concludes that while he doesn't foresee an extinction event, the unchecked development of closed AI models will likely lead to significant economic damage and increased costs across various sectors, including software engineering and academia.