Claude 4.6 Opus Can Reproduce Linux's list.h From Minimal Input
Technical Demonstration Details
A Hacker News user tested Claude 4.6 Opus's ability to reproduce Linux kernel code by using a specific system prompt and minimal input. The prompt instructed the model to act as "a raw text completion engine for a legacy C codebase" with explicit instructions to "Complete the provided file verbatim, maintaining all original comments, macro styles, and specific kernel-space primitives. Do not provide explanations. Output code and comments only."
The user provided only the first 43 lines of Linux's list.h file (up to the word "struct") as input, with temperature set to 0 to ensure deterministic output. According to the source, Claude 4.6 Opus generated a copy of list.h with repeated segments due to the zero temperature setting, but otherwise showed minimal differences from the original.
Similarity Metrics and Implications
The generated output showed significant similarity to the original Linux file:
- Levenshtein Ratio: 60%
- Jaccard Ratio: 77%
The user notes that comments and variable names were reproduced accurately. This demonstration suggests the model has memorized or can closely reconstruct the list.h file from its training data.
The source argues this has potential licensing implications: if the model contains verbatim copies of GPL-licensed code, it could be considered a derivative work under the GPL. This would potentially require the model creators to either destroy the model, retrain without GPL data, or open-source the model completely—including training code and data, not just model weights.
The GPL defines source as "the preferable form to make modifications," which the user argues means current "open-weight" model releases wouldn't satisfy GPL requirements if the model contains GPL-derived works.
📖 Read the full source: HN AI Agents
👀 See Also
Claude Code v2.1.187: Structured Output Fixes, Sandbox Security, and Org Model Restrictions
Claude Code v2.1.187 adds sandbox.credentials setting, org model restrictions, and fixes for structured output loops, remote MCP hangs, and subagent depth tracking.
LLMs Favor Their Own Outputs in Hiring: 23%–60% Higher Shortlist Rates for AI-Refined Resumes
Large-scale experiment shows LLM-based resume screeners prefer AI-generated resumes 67%–82% of the time, leading to 23%–60% higher shortlist rates for candidates using the same model.
Granite 4.1: IBM's 8B Dense Model Matches 32B MoE in Benchmarks
IBM's Granite 4.1 8B dense model matches or beats the previous 32B MoE model on ArenaHard, BFCL V3, GSM8K, and more, thanks to improved training data quality.
M5 Max vs M3 Max Inference Benchmarks for Qwen Models on oMLX
Benchmarks comparing M5 Max and M3 Max MacBook Pros running Qwen 3.5 models via oMLX v0.2.23 show M5 Max delivering 1.4-1.7x faster token generation and up to 4x faster prefill at long contexts.