Skip to content

Instantly share code, notes, and snippets.

View whit3rabbit's full-sized avatar

whit3rabbit whit3rabbit

55 followers · 9 following
View GitHub Profile
@whit3rabbit
whit3rabbit / guess_claude_watermark.md
Last active August 15, 2026 16:25
Guessing what watermarking might look like in Claude

Model-Level Watermarking in Large Language Models: Anthropic's Claude, Keyed Sampling Provenance, and Watermark Propagation

Summary

In August 2026, Anthropic disclosed how its Claude models watermark generated text. This paper explains the disclosed system, what it does and does not prove, and how to study it experimentally. The main points:

  1. Claude's watermark is a decoding-time, keyed pseudorandom sampling watermark. It is a variant of the SynthID-Text method published by Google DeepMind (Nature, 2024), in the family of approaches that began with Aaronson's 2022 proposal.
  2. The watermark changes only the source of randomness used to pick among candidate tokens during sampling. Nothing is added to the text, there are no hidden characters, no extra tokens are generated, and Anthropic reports no measurable impact on output quality, speed, or cost.
  3. Detection requires a secret key held by the provider, and it yields a likelihood that Claude was involved in producing a passage, not a binary auth
@whit3rabbit
whit3rabbit / ZAI_CC.txt
Last active April 14, 2026 01:41
zai with claude code
# Get coding plan
https://z.ai/subscribe
You then go generate an API key.
# use an isolated version of claude code with z.ai support
# https://github.com/numman-ali/cc-mirror
# Z.ai (GLM-5/4.7/4.5-Air)
npx cc-mirror quick --provider zai --api-key "$Z_AI_API_KEY_HERE"
@whit3rabbit
whit3rabbit / CLAUDE.md
Last active May 10, 2026 00:54
My ~/.claude/CLAUDE.md

Defaults

  • macOS, en-US.
  • Be direct and critical. Say when the user is wrong.
  • Prefer security, correctness, then speed.
  • Keep changes small, reversible, and easy to review.
  • Do not present guesses as verified facts.
  • Use st instead of ripgrep/grep

Style

@whit3rabbit
whit3rabbit / APKM.md
Last active October 13, 2025 01:03
Instructions on how to convert/inspect APKM file
@whit3rabbit
whit3rabbit / package.json
Last active September 18, 2025 23:22
Claude Code Deobfuscated v1.0.77 (processed by LLM)
{
"name": "@anthropic-ai/claudie-code",
"version": "1.0.117",
"description": "Anthropic's official CLI for Claude, designed to help with software engineering tasks. This is a best guess deobfuscation.",
"type": "module",
"main": "dist/entrypoints/cli.js",
"bin": {
"claude": "dist/entrypoints/cli.js"
},
"files": [
@whit3rabbit
whit3rabbit / valacheatsheet.md
Last active July 20, 2025 23:17
valacheatsheet.md

The Vala with GTK4 Cheat Sheet

1. Installation & Project Setup

Installation

  • Fedora: sudo dnf install vala gtk4-devel
  • Debian/Ubuntu: sudo apt install valac libgtk-4-dev
  • Arch Linux: sudo pacman -S vala gtk4
@whit3rabbit
whit3rabbit / old_python_kali.py
Created March 21, 2025 17:21
Install older python versions in kali
# It's not recommended to install other repos unless you have to
#
sudo apt-get update
sudo apt install python3-launchpadlib software-properties-common
sudo add-apt-repository ppa:deadsnakes/ppa
sudo sed -i 's|deb https://ppa.launchpadcontent.net/deadsnakes/ppa/ubuntu/ kali-rolling main|deb https://ppa.launchpadcontent.net/deadsnakes/ppa/ubuntu/ noble main|' /etc/apt/sources.list.d/deadsnakes-ubuntu-ppa-kali-rolling.list
sudo apt-get update
sudo apt install python3.10 python3.10-venv
@whit3rabbit
whit3rabbit / find_token.c
Last active September 11, 2024 20:01
Find token
#include <windows.h>
#include <stdio.h>
#include <stdlib.h>
#include <psapi.h>
#pragma comment(lib, "psapi.lib")
#define NT_SUCCESS(Status) (((NTSTATUS)(Status)) >= 0)
#define STATUS_INFO_LENGTH_MISMATCH 0xC0000004
typedef LONG NTSTATUS;
@whit3rabbit
whit3rabbit / hide_process_name.c
Last active October 4, 2024 16:25
hide_process_name.c
#define _GNU_SOURCE
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <sys/prctl.h>
#include <sys/mman.h>
#include <errno.h>
#include <dirent.h>
#include <sys/mount.h>
#include <sys/stat.h>
@whit3rabbit
whit3rabbit / golden_gate_gpt2.py
Created May 27, 2024 18:58
Trying to understand https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html
# !pip install torch transformers scikit-learn umap-learn matplotlib datasets joblib pandas zstandard
'''
Code outline:
1. Loading the pre-trained GPT-2 model and tokenizer.
2. Loading or downloading the dataset and saving it to disk with compression.
3. Collecting and normalizing activations from the middle layer of the model.
4. Training multiple SAEs with different feature sizes and saving them to disk.
5. Finding the feature that responds to the phrase "Golden Gate Bridge" in each SAE.