The Matthew Brett Controversy - What Actually Happened

The Context

A NumPy pull request (#30828) was suspected by some in the NumPy mailing list discussion of potentially using

AI-generated code. Matthew Brett decided to investigate this using Google's Gemini AI as a "copyright red team."

Matthew Brett's Action

He posted a GitHub comment where he:

1. Ran Gemini AI over the PR with a specific prompt asking it to act as a "copyright red team"
2. Published Gemini's full response (now deleted, link to gist) which claimed:
• The code was "likely AI-generated" (specifically suggesting Claude 3.5 Sonnet or GPT-4o)
• The constant 384 was "suspicious" and "AI-style"
• The code might be derivative of JAX (Apache 2.0) or Apache Arrow
• Suggested potential "license contamination" from GPL-licensed libraries
• Concluded with a recommendation that maintainers ask the PR author to explain the source of the 384 constant

Why People Were Angry at Matthew Brett

1. The method was seen as inappropriate:

• Using AI to analyze code for AI-generated content creates a circular, unreliable detection method
• The "Red Team" framing was seen as adversarial and accusatory toward a fellow contributor

2. The specific allegations were damaging:

• Gemini's response directly accused the PR author (@mdrdope) of potentially:
• Using AI without disclosure
• Copying from JAX without attribution
• Risking GPL "license contamination"
• These are serious accusations in open source that could damage a contributor's reputation

3. The evidence was flimsy:

• The "suspicious constant 384" was just a heuristic - not actual proof
• The AI-generated analysis was itself speculative ("likely," "suggests," "non-zero risk")
• The comment style analysis ("pedantic style characteristic of Claude 3.5 Sonnet") is pseudoscientific

4. It put the PR author in an impossible position:

• Asking someone to "prove they didn't use AI" is like asking someone to prove a negative
• The comment created pressure for @mdrdope to defend their work against algorithmic suspicion

5. Robert Kern's specific objection:

• Found the AI-generated text "low-quality and misleading"
• The comment was seen as "rude" to the PR author
• The copyright violation suggestions were "unfounded" upon further investigation

The Outcome

Matthew Brett deleted the original comment "in the interests of peace" and reproduced it in the gist for the mailing list
discussion to see what the controversy was about.

---

Key Takeaway

The anger wasn't about Matthew Brett's caution regarding AI in general (as discussed in the mailing list thread), but specifically
about this GitHub comment where he used AI to accuse a specific contributor of potential copyright violations and undisclosed
AI use based on speculative pattern-matching - particularly the "suspicious constant 384" analysis which sounded authoritative
but was essentially algorithmic phrenology.