Gebru and Bender: this summer's AI breakthrough hype is mostly marketing

MIT Tech Review AI · rss · 2026-09-22

Timnit Gebru (DAIR) and Emily Bender (UW) dissect this summer's AI hype cycle in MIT Tech Review:

Hacking incidents reframed: cybersecurity experts say the "rogue model" stories were really OpenAI negligence of basic security practices — the framing helps companies dodge accountability.
Math breakthroughs challenged: mathematicians found OpenAI's Astra claims far less novel than advertised, accusing the company of research misconduct; NYU's Tristan Buckmaster alleged OpenAI stole others' work.
Why math and coding? Both sell the "all-knowing machine" narrative and are cheap to train on since outputs are automatically verifiable.
Ideological critique: "incipient superintelligence" claims rest on transhumanist ideology, not science, while real harms — data centers' climate impact, asthma, electricity bills, water use — get dismissed as "distractions."
The authors echo hundreds of mathematicians urging policymakers to consult independent experts rather than press releases.

More from AGI Musings