Gebru, Bender: AI hype summer masks marketing, not science

The authors situate this summer's events within a months-long sequence: Anthropic's late-April Claude Mythos claim about finding software vulnerabilities, the OpenAI–Hugging Face hacking incident, similar disclosures by Anthropic and Meta, then dueling mathematical-breakthrough claims from Anthropic and OpenAI. They note that each incident drew breathless press coverage repeating the companies' anthropomorphizing framings, while the more measured expert assessments that followed received far less attention. The piece also highlights engineer Jacob Coxon's viral departure from Anthropic, where he claimed the company and OpenAI are "racing straight towards self-improving superintelligence," as an example of rhetoric that gets amplified.

The article argues that math and computer programming receive disproportionate attention partly because answers in those fields can be verified without paid annotators, making them easier to tune systems for and to market as pinnacle human achievements. Mathematicians, the authors write, have warned against this corporate use of their field — a statement signed by hundreds calls for policymakers to consult experts rather than press releases. Gebru and Bender extend that call, arguing that 'superintelligence' framing ascribes agency to products and helps companies evade accountability, and that policy decisions should follow independent expert input rather than corporate urgency. The stakes, as they frame it, hinge on whether policymakers and the public hold onto skepticism the next time the hype cycle arrives.