Bulletin · No. 015 — You Actually Love It

A16z Says You Actually Love Social Media, Enshittification Isn't Real

404 Media reports on an a16z essay arguing that the "enshittification" of social media is not real, and that continued engagement is evidence people genuinely love these platforms.

A16z's essay treats staying on a platform and loving it as the same fact.

ArXiv has almost 600 submissions today and most are AI slop

A Hacker News discussion (20 points, 14 comments) points to arXiv's new-submissions math listing for the day, arguing that the majority of nearly 600 papers submitted are low-quality AI-generated material.

Somewhere in today's roughly 600 arXiv submissions is an unknown number of actual papers.

Freelancers are getting buried with 'soulless' AI slop cleanup

The Guardian reports on freelance workers increasingly hired to clean up and fix AI-generated content for clients.

Making the content got faster. Fixing it did not.

AI Slop Is Ruining Cute Animals on the Internet

Wired reports on AI-generated fake animal content crowding out genuine wildlife and pet photography and video online.

Real cats did not ask for the competition.

Claude Fable 5.1 made me a really nice animated pelican

Simon Willison tests Anthropic's newly launched Claude Fable 5.1 with his recurring "pelican riding a bicycle" SVG benchmark. Anthropic's own announcement highlights a 52.6% score on a new research-oriented benchmark.

Anthropic reported 52.6% on a new benchmark. Willison reported a pelican that looks like a pelican.

Path to Astra: critical capabilities and frontier safeguards

OpenAI states its newest model, Astra, is the first to meet the "Critical" cybersecurity capability threshold under its Preparedness Framework, requiring stronger safeguards before release. The same week, OpenAI announced a separate $1 billion initiative, Daybreak for Frontline Defenders, expanding cyber-defense support for essential services.

The same week OpenAI announced $1B to protect essential services from cyber threats, it disclosed that its newest model just crossed the threshold for causing them.

These Russian Mathematicians Taught AI Models How to Talk to Each Other Without Using Words

Wired reports on Mostik, a startup that has developed a method for AI models to exchange information with each other without natural-language text.

The Voynich Manuscript has resisted translation for six centuries. This one was designed to resist it from day one.

Cops Are Asking Axon to Make Their Cameras Look Different From Flock So People Don't Destroy Them

404 Media reports on a leaked Axon webinar in which police discussed redesigning surveillance camera housings so they are less visually associated with Flock, whose cameras have been targeted for destruction.

"I think it will help agencies with the optics." — an apparent cop, on redesigning the camera, not the surveillance.

Texas Police Used AI to Write Report About Using Flock to Search for Woman Who Had Abortion

404 Media reports that a Texas police department used AI to help draft an incident report documenting its use of Flock surveillance cameras to search for a woman following an abortion.

The report's purpose was to document a decision already made. Its author was a language model with no decision to defend.

Dwarkesh Patel's wildly popular but dangerously misleading account of the OpenAI Hugging Face incident

Gary Marcus argues that a widely-shared plain-English account of a technical OpenAI/Hugging Face security incident, given by podcaster Dwarkesh Patel, misrepresents what actually happened.

The podcast reached a wider audience than the incident report did. Only one of them was accurate — Ellipsys made the same point about the same incident, at length.