Appearing Productive in the Workplace

GPT‑5.5 Instant

Accelerating Gemma 4: faster inference with multi-token prediction drafters

Three Inverse Laws of AI

Lessons for Agentic Coding: What should we do when code is cheap?

Train Your Own LLM from Scratch

Let's talk about LLMs

LLMs Are Not a Higher Level of Abstraction

Specsmaxxing – On overcoming AI psychosis, and why I write specs in YAML

Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a coding challenge

The agent harness belongs outside the sandbox

AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights

Refusal in Language Models Is Mediated by a Single Direction

The gay jailbreak technique

DeepSeek V4–almost on the frontier, a fraction of the price

The More Young People Use AI, the More They Hate It

Granite 4.1: IBM's 8B Model Matching 32B MoE

Where the goblins came from

Alignment whack-a-mole: Finetuning activates recall of copyrighted books in LLMs

The Zig project's rationale for their firm anti-AI contribution policy

Vera: a programming language designed for machines to write

Why AI companies want you to be afraid of them

He asked AI to count carbs 27000 times. It couldn't give the same answer twice

We decreased our LLM costs with Opus

I won a championship that doesn't exist

Silicon Valley has forgotten what normal people want

Running local LLMs offline on a ten-hour flight

TurboQuant: A first-principles walkthrough

Eden AI – European Alternative to OpenRouter

Amateur armed with ChatGPT solves an Erdős problem

Lambda Calculus Benchmark for AI

Google to invest up to $40B in Anthropic in cash and compute

OpenAI releases GPT-5.5 and GPT-5.5 Pro in the API

Technical, cognitive, and intent debt

Kernel code removals driven by LLM-created security reports

Types and Neural Networks

Even 'uncensored' models can't say what they want

We got 207 tok/s with Qwen3.5-27B on an RTX 3090

GitHub's Fake Star Economy

Claude Token Counter, now with model comparisons

Changes in the system prompt between Claude Opus 4.6 and 4.7

Claude Opus 4.7 costs 20–30% more per session

George Orwell Predicted the Rise of "AI Slop" in Nineteen Eighty-Four

Android CLI: Build Android apps 3x faster using any agent

Claude Opus 4.7 Model Card

Claude Opus 4.7

The future of everything is lies, I guess: Where do we go from here?

AI cybersecurity is not proof of work

SDL bans AI-written commits

The local LLM ecosystem doesn’t need Ollama