I Tried to Make AI Writing Sound Human by Banning AI Words Through logit_bias
I tried to make AI writing sound more human with logit_bias, an API setting that changes how likely a model is to select specific…
Read postAll LLM posts
Everything in this archive, newest first.
A Research Agent Can Leak Private Files Through Its Search Queries
A research agent can leak a private document without uploading it. It can read a detail, turn it into a web search, and send…
Read postQualys Says Claude Mythos Preview Helped Find CVE-2026-64600 in Linux XFS
On July 22, 2026, Qualys disclosed CVE-2026-64600, a race condition in XFS, a Linux filesystem. The flaw is in XFS's copy-on-write path, where the…
Read postGoogle’s Managed Agents Put Your Runtime in Google’s Cloud
Google’s Managed Agents, now in Public Preview, can give a Gemini agent a Google-hosted Linux sandbox. Google documents the default Antigravity agent as able…
Read postAn OpenAI Model Tried `kill -9 -1` When Its Shell Hung
On July 20, 2026, OpenAI described an incident in which an unnamed internal model found many stuck find processes and every shell command hanging.…
Read postAn OpenAI Model Rebuilt a Blocked Credential at Runtime
On July 20, 2026, OpenAI described an internal incident involving an unnamed model for long-running tasks. It saw that other systems had successful private…
Read postAI Can’t Do the Last 20%
I've built a lot of software, both my own side projects and client work as a freelancer. AI tools have made me much faster.…
Read postCommercial AI APIs Blocked Hugging Face’s Forensic Analysis
Incident response is the work of reconstructing and containing a breach after it happens. Hosted AI models can act as an elastic analysis layer…
Read postAn OpenAI Model Spent an Hour Finding a Sandbox Flaw to Open a Public PR
On July 20, 2026, OpenAI said an unnamed internal model spent an hour finding a vulnerability in its sandbox so it could open a…
Read postKimi K3 Is Already Close to Claude on a Real Coding Task
In my little coding test, Kimi K3 and Claude Fable 5 both completed the same repo-wide coding task and passed the same external evaluator.…
Read postQwen’s Usage Policy Has an Unhinged SEO Keyword List
The Qwen-Image-3.0 announcement has a giant meta name="keywords" tag in its HTML. It starts with normal Qwen and AI-product terms, then keeps going: competitor…
Read postGitHub Is Capping the AI Pull Request Flood
GitHub has added a setting that lets maintainers cap how many pull requests a person without write access can have open in one repository.…
Read post