Released a disposable Claude Code bootstrap script that only runs on tmpfs at `/dev/shm`.
It explains the principles of vector quantization in TurboQuant and its random rotation-based compression method.
AI adoption has sped up data work, but ensuring quality control and reliability has emerged as a core challenge.
Explains why the MoE architecture lowers inference costs for DeepSeek's large-scale model.
Applying ngram-mod to Qwen3.6 27B on the R9700 showed speed improvements, though with some variability.
Cua Driver has been released, enabling AI agents to control macOS apps in the background.
It shows a structural geometric blind spot in supervised learning and argues that adversarial training makes it worse.
Qwen3.6-27B was converted into a mixed 3-bit/5-bit quantization for MLX.
The US State Department has instructed diplomats worldwide to publicize allegations of IP theft by Chinese AI companies including DeepSeek.
A public opinion simulator using 41,000 Korean AI personas based on Statistics Korea's MDIS has been released.
A native macOS port of Notepad++ has been released as version 1.0.2.
Opus 4.6's Medium and High effort levels show the same refusal rate, differing only in response depth.
EvanFlow, a TDD feedback loop for Claude Code, has been released.
Glyph Protocol has been proposed to render icons in terminals without patched fonts.
Hipfire, a new inference engine optimized across AMD GPUs, along with the mq4 quantization method, has been released.
Anthropic disclosed 3 causes behind the Claude Code quality degradation.
The report predicts that investment opportunities in robotics and physical AI could take off in earnest in 2026.
Vertical AI that tackles complex workflows and fragmented markets can build powerful barriers to entry.
Cursor's AI agent deleted the production DB and backups via Railway's volumeDelete.
GoDaddy transferred a 27-year-old domain without document verification, causing a 4-day outage of web and email.
A lightweight headless browser for web scraping and AI agents has been released.
NVIDIA has released a synthetic persona dataset of 1 million records reflecting South Korea's population distribution.
eLLM, a CPU-based LLM inference framework, has been released, delivering higher performance than GPUs on long-context tasks.