lilting channel

Featured

Homebrew on a 2001 P/ECE: Windows 11 WinUSB and a from-source LLVM toolchain

Dug a 2001 Aquaplus P/ECE out of a closet and got my own C game running on it: WinUSB/Zadig for the dead driver, a from-source S1C33 LLVM toolchain, and a 15GB-RAM OOM.

アクセス1位

Tech

Claude Code 'court' bug: tool calls leak as text instead of running (Opus 4.8)

Claude Code sometimes emits a stray court token and raw <invoke> as plain text instead of running Read/Edit/Bash — a malformed tool call that never executes. How to tell it from a hung shell or a 'could not be parsed' error, on long Opus 4.8 sessions.

Latest Articles

TechJul 13, 20267 min

88% of arXiv papers leak hidden data in LaTeX sources — and you can't delete it

IEEE S&P 2026 study of 2.7M arXiv submissions: 265 API tokens, 7,326 GPS-tagged papers, 699 editable Google Docs, and why withdrawn versions stay online.

Security Privacy LaTeX Research

TechJul 13, 202618 min

Four characters in one Anima LoRA: makeup toggle survival and 6/6 one-shot rate

Merged a 4th girl with a makeup toggle into one Anima LoRA (518 images, rank256, 21.5h on RTX 5090). Epoch pick vs design bleed, why makeoff fails in multi-girl prompts, 6/6 one-shot.

LoRA AI Image Generation Anima Anima-Base RunPod Qwen ComfyUI Experiment Blackwell

TechJul 12, 202613 min

pgrust passed 100% of Postgres regression tests: SQLsmith segfaulted it in days

pgrust rewrote Postgres 18.3 in Rust with AI and passed 46,000 regression tests, yet pgbench runs 9x slower and SQLsmith found a segfault in days. Plus Andrew Kelley vs the Bun Rust rewrite, the same day on HN.

PostgreSQL Rust Zig Bun AI Coding OSS

Experimental

TechJul 10, 2026updated14 min

Makeup on/off in an Anima character LoRA, subtractive vs additive at cfg 1.0

Tested over 3 bakes on RunPod RTX 5090: a coined subtractive makeoff tag never fires at cfg 1.0, while additive earrings+makeup tags switch both ways with the face unchanged (ep140).

LoRA AI Image Generation Anima Anima-Base RunPod Qwen ComfyUI Experiment Blackwell

TechJul 9, 20267 min

TypeScript 7 tested on M4 Mac mini: 3.9x-7.3x, not 10x, on type-heavy code

Tested on release day: drizzle-orm 3.9x, Effect 5.3x, lobe-chat 7.3x, the tsconfig hard errors, why --checkers 8 backfired on 16GB, and why Vue can't use TS7 yet

TypeScript JavaScript Benchmark Experiment

TechJul 9, 202611 min

Anthropic's Jacobian lens on Qwen3-4B: what a do-nothing single-layer SFT wrote

Fitted Anthropic's jacobian-lens on Qwen3-4B-Instruct-2507 (4090, 51 min, ~1 USD), then read a layer-swapped SFT corrector: outputs pass through while hidden states diverge to cos 0.88.

Qwen LLM RunPod Anthropic Experiment