A Hacker News thread on proving human authorship of text drew 31 points and 28 comments after the original post appeared at gjtorikian.online.
The discussion centers on practical signals that distinguish human writing from large language model output without relying on unreliable detectors.
Core Techniques Discussed
Commenters outlined two main approaches. Cryptographic signing attaches a private-key signature to a document hash before publication. Stylistic markers include deliberate inconsistencies in sentence length, personal anecdotes, and references to physical context that current models rarely replicate consistently.
Both methods require the author to act before any dispute arises.
Measured Effectiveness
Early testers in the thread reported that signed documents pass verification in under two seconds using standard tools such as GPG. Unsigned text relies on manual review; one commenter noted that stylistic checks caught AI-generated passages in roughly 70 percent of sampled cases from recent blog posts.
No large-scale benchmark numbers were shared.
How to Apply Signing Today
Generate an Ed25519 key pair with ssh-keygen -t ed25519.
Hash the final document and sign it: gpg --sign --armor document.txt.
Publish both the signed file and the public key on a personal site or Git repository.
Readers verify with gpg --verify in any terminal.
Tradeoffs Reported
- Signing adds one extra file and a public key step; most readers skip verification unless stakes are high.
- Stylistic signals work only when the author maintains consistent habits across years of writing.
- Detectors based on token probability remain easy to defeat with light editing.
Comparison With Common Alternatives
| Method | Verification Time | Requires Pre-Planning | False Positive Rate |
|---|---|---|---|
| GPG signature | <2 s | Yes | Near zero |
| Watermarking APIs | Instant | Yes (model dependent) | Unknown |
| Commercial detectors | 5–30 s | No | 15–40 % reported |
| Manual stylistic review | Minutes | No | Author dependent |
Who Benefits Most
Developers publishing technical notes or researchers releasing preprints gain immediate value from signing. Writers who publish only casual social media posts see little return on the added workflow.
Teams already using version control can integrate signing into existing commit hooks with minimal friction.
Practical Verdict
Signed documents provide the only deterministic proof currently available; stylistic analysis serves as a useful secondary signal when signatures are absent.
Bottom line: Cryptographic signing combined with consistent personal style offers the strongest practical defense against authorship disputes today.
The thread shows growing interest in lightweight signing workflows as model output volume increases.
Top comments (0)