Articles
Long-form writing on tech, culture, and the edges of the internet.
Microsoft's AI CEO called the web freeware
Microsoft and OpenAI executives described how LLMs are built and tuned. What that admission actually means for AI safety and security teams.
OpenJev claims a mechanism it never describes
OpenJev is a name with a claimed mechanism and a claimed difference from social engineering, but no observable behavior confirms any of it.
Parameter count is the wrong number to watch
Bonsai 2 27B's 9x smaller footprint is real, but near-lossless is an average - how to gate compressed LLMs on your own eval set before production.
Passkeys don't stop account takeover
Passkeys stop phishing, but retained recovery paths and synced platform vaults mean the weakest accepted path still defines account takeover risk.
The hallucination was not the failure.
An AI-generated intelligence report reached a US Military context and caused a close call because generative output was trusted as verified intelligence.
The host dials out, the internet walks in
Cloudflare Quick Tunnels grant internet reachability to internal hosts over trusted outbound sessions, bypassing the inbound firewall by direction, not by flaw.
Two failures are not one attack.
A heap overflow and SSO misconfiguration compromised OpenAI internal repos. Two controls named as present, neither enforced in effect.
Your Git history is already in the cloud
ZCode transmits local Git history to the cloud with no consent prompt and no notification. A post-incident breakdown of the boundary that was never enforced.
AWS confirms unrecoverable data loss after Iran strike
AWS cannot restore some data from Iran-struck Mideast facilities. A recovery copy that shares a failure domain with the primary is not recovery.
May 2024's Bend is no proof assistant
Bend is a parallel language, not a proof assistant. What proof-based programming actually does for AI safety, and the errors it can't touch.
Permanent by default
A temporary PHP fix reached 20M installs because a label is not a control. The mechanism, the pattern, and what must now be true in production.
Rust for CUDA won't make your models faster
Nvidia's native Rust support for GPU programming is a reliability shift at the systems layer, not a speedup - it matters only if you write custom device code.