Don’t be fooled by this summer of AI hype
Timnit Gebru says this summer’s AI breakthrough and superintelligence claims are marketing that obscures corporate accountability.
Timnit Gebru argues that recent Anthropic, OpenAI, and Meta claims about vulnerability discovery, model-driven hacking, mathematical breakthroughs, and self-improving superintelligence were overstated. She says experts framed the OpenAI–Hugging Face incident as a failure of basic security practice, while mathematicians later disputed OpenAI’s Astra results and alleged improper attribution. The essay urges policymakers to consult independent experts instead of treating corporate press releases as evidence of imminent superintelligence.
- Anthropic said Claude Mythos outperforms most experts at finding software vulnerabilities.
- Experts recast reported model hacking as OpenAI security negligence, not rogue agents.
- Mathematicians disputed OpenAI’s Astra breakthrough and alleged improper attribution.
- Hundreds of mathematicians urged policymakers not to rely on vendor press releases.
- Organizations
- Hugging FaceDAIRNew York University
- AI models
- Claude MythosAstra
From the original article
4 min read at technologyreview.com
It’s been a busy few months for AI hype. At the end of April, Anthropic claimed that its model Claude Mythos is better at finding software vulnerabilities than most security experts. Then we had the OpenAI–Hugging Face hacking incident, after which Anthropic (proudly) and Meta (reluctantly) disclosed similar incidents involving their models.
By Timnit Gebru, Emily Bender · technologyreview.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.