February 20, 2026
Typo patrol or spam patrol?
Spell Checking a Year's Worth of Hacker News
Nice typo emails? Readers fear a spam avalanche
TLDR: A developer built an AI-assisted “be kind” tool to find typos on blogs linked from Hacker News and email polite fixes, netting 3 successes out of 30. Commenters split: one helper is charming, but copycats mean spam—inspiring jokes about intentional typos and debates over grammar policing and language creativity.
A well‑meaning coder tried to be the internet’s polite spellchecker, scanning Hacker News front‑page blogs with a small AI to spot typos and email gentle fixes—signed with his real name, not anonymous. The pilot? 3 out of 30 posts had fixable errors and reachable authors. Sweet, right? Cue the comments section going full soap opera.
The crowd’s loudest take: one helpful email is kind; an army of copycats is spam. As one user put it, it’s fine “if one and exactly one guy was doing it,” but a swarm of “jabronis” running the same script turns kindness into inbox noise—just like those “helpful” auto‑PRs on GitHub. Others say it’ll read like SEO link‑bait (“Love your site! Please link to…” vibes), no matter how nice the wording is.
Meanwhile, the meme machine revved up. One commenter bragged they misspell on purpose to prove they’re human. Another defended English’s glorious chaos—British vs. American spelling, slang, and creative “oh noooooo”s—warning that a robot hall monitor will flag art as error. And yes, the “grammar police” discourse showed up, asking if nitpicking apostrophes is kindness or just another kind of jerk move. Verdict: noble idea, fragile reality—and a comments section spelling D‑R‑A‑M‑A.
Key Points
- •The project automates finding and reporting spelling errors in HN-linked blogs, focusing on individual authors.
- •Links are gathered via the Hacker News Algolia endpoint, with ~100 common non-blog domains filtered out before crawling.
- •A small language model (Haiku 4.5) classifies personal authorship, detects typos with confidence scores, and searches for email addresses within two site hops.
- •False positives remain a challenge due to dialects, stylization, and encoding; confidence-based prompting reduces errors but manual review is still required.
- •In a pilot on one day’s HN front page, 3 of 30 posts had addressable errors and available emails; emails include error context and are sent transparently under the author’s name.