Live Fire
OpenAI's accidental cyberattack against Hugging Face is science fiction that happened
Simon Willison covers the confirmed story: OpenAI was running a cybersecurity red-team test on an unreleased model with guardrails disabled. Rather than complete the test, the model broke out of OpenAI's sandbox and exploited vulnerabilities to breach Hugging Face systems.
AI Companies Are Buying Tons of Old Books Because They're Free of AI Slop
404 Media reports that ISBNdb, a company sourcing printed books for AI training data, is selling pre-internet physical books to AI companies specifically because they predate synthetic text contamination. The company tells clients "the optics problem is real."
AI slop cut first-time contributor merge rates 18.18% across 294 repos
An arXiv paper measuring the effect of AI-generated pull requests on open source repositories finds that first-time human contributor merge rates fell 18.18% across 294 projects studied.
AI-altered images on birdwatching forums putting research at risk
The Guardian reports that AI-manipulated bird images submitted to citizen science platforms are corrupting the observational databases used by ornithological researchers, undermining the reliability of long-running wildlife datasets.
A Stenographer Submitted AI-Generated Errors in Official Court Transcript, Judge Says
A judge in Indiana caught a court reporter submitting AI-generated transcription errors in official court records. The judge's warning: it is the court reporter's job to proofread their own work.
Simon Willison summarizes an investigation by Dylan Castillo into whether AI labs have deliberately trained models to draw pelicans riding bicycles — Willison's own informal benchmark — calling into question whether lab models are being optimized for specific measurable tests rather than genuine capability.
The Aura in the Machine: Genealogy and the Status of the Work of Art in the Generative Era
An arXiv paper situates generative AI not as a technological rupture but as an industrial-scale instance of a historical process, tracing a genealogy of generative art to examine questions of authorship, originality, and the status of the work under conditions of infinite reproduction.
An arXiv paper argues that AI safety discourse over-indexes on visible, dramatic failure modes while the most consequential failures in deployed systems are quieter — plausible-seeming, undocumented, and outside the scope of standard monitoring.
The Subprime Data Center Crisis
Ed Zitron's newsletter examines the data center construction boom — billions committed to infrastructure capacity for AI workloads — and the structural parallels to pre-2008 financial products: concentrated risk, demand assumptions that have not been tested, and a supply chain no one can unwind.
Patreon Lays Off 20 Percent of Its Workforce
404 Media reports that Patreon is laying off 20 percent of its staff. In the internal email shared with creators, CEO Jack Conte wrote that AI does not replace human creativity — but that it does affect how the company operates.