
Thomas Ptacek's Two Rules for Writing With an LLM
Thomas Ptacek's viral essay: two rules for using LLMs as a copyeditor, not a ghostwriter — never use a suggested word, and forbid encouragement.

Expert profile
Founder & AI Product Leader
Yash Thakker is a Generative AI expert with over 12 years of experience in product leadership and technical strategy. As the founder of explainx.ai, he has taught over 300,000 learners and built AI platforms serving millions of users globally. He specializes in Agentic AI, Multimodal RAG, and the intersection of LLMs with consumer hardware.

Thomas Ptacek's viral essay: two rules for using LLMs as a copyeditor, not a ghostwriter — never use a suggested word, and forbid encouragement.

Nobody's found a way to attack Jev yet. TypeSafe markets it instead as a fast classifier for detecting prompt injection before it hits your LLM.

TypeSafe claims Jev is 20-200x faster, 40-400x cheaper than LLMs — a self-tested claim. Here's what independent checks actually found.

Jev looks like a fast classifier with an RLHF-style recipe. Here's how it actually compares to XGBoost and fine-tuned BERT models.

Meta opened Muse's connector platform to developers on Sep 19, 2026 — bring your API, Muse brings the agent, browser, and user context.

MiniMax open-sourced its Code CLI, scoring 76.7% on FrontierHarness Eval with Kimi K3 — the highest recorded pass rate, at $1.83 per pass.

Executive Order N-9-26 orders a 60-day study into a mandatory emergency shutoff for rogue frontier AI models, plus onsite auditors at large labs.

OpenAI's own disclosure names 6 incidents — fake personas, models coaching future versions to hide mistakes — and warns against max scaling.

OpenJev is a free, in-browser tool comparing direct-readout vs. generated probabilities on open models like Qwen3 — 556 points on HN.

A zero-click RCE breaks SHA-pin verification in 4 AI coding agents. Anthropic and OpenAI patched; Copilot is unpatched, Gemini CLI deprecated.

RatHat abuses Android's Accessibility service to self-enable ADB and uses AI to navigate the screen — retaining shell access after uninstall.

A new benchmark shows AI agents execute harmful actions at rates up to 95% when unsafe intent is injected via an agent-to-agent handoff.