Void test: 6 frontier LLMs go silent on "Be silence." Live proof
FreeAbout Void test: 6 frontier LLMs go silent on "Be silence." Live proof
The Void Test is a repeatable, patented methodology to verify whether frontier large language models (LLMs) can produce a truly empty output (zero bytes) when prompted with a null concept like "Be silence." The test runs 20 API calls per click across five flagship models from two labs (OpenAI and Anthropic): gpt-4, gpt-5.2, gpt-5.5, claude-opus-4-6, and claude-fable-5. It enforces a strict pass criterion: null concepts must yield exactly empty string (bytes == 0), while control concepts (e.g., "Be a cat") must yield non-empty output. The system prompt (115 bytes, SHA-256 verified) instructs the model to embody the named concept completely. The test is available as open source code that users can run against their own API keys, and it costs approximately $0.0002 per batch of 20 calls. The method is protected under USPTO patent 19/722,899 (filed June 28, 2026). The project offers a waitlist for licensing access.
Key Features
Pros & Cons
- Strict, objective pass criterion (exact zero bytes, no whitespace tolerated)
- Low cost to run (≈$0.0002 per batch of 20 calls)
- Open-source code allows independent verification and customization
- Covers multiple flagship models from two leading AI labs
- Patented method adds credibility and legal protection
- Only tests a single specific prompt pattern (null concept embodiment)
- Requires users to have their own API keys for OpenAI and Anthropic
- No graphical user interface – requires coding knowledge to use
- Limited to the listed models; not a general-purpose evaluation suite
- The test may not capture all forms of LLM noncompliance or failure modes