Project ideas from Hacker News discussions.

Show HN: Qwen3-4B with the AI writing tics trained out

📝 Discussion Summary (Click to expand)

Theme 1 – Distracting UI (cursors) and the need to hide it

“What's with the very distracting cursors of others?” – anilgulecha
“Exactly. I was interested in the article but I don’t have time for that nonsense.” – mullenba
“Agreed. In the upper right corner there's a 'hide' button that I immediately clicked. Then read (ok, skim read) the article.” – markgavalda
“Just because you can do a thing, doesn't mean you should.” – HumblyTossed

Theme 2 – AI‑generated text tends toward a uniform, recognizable “slop” style

“Models have proclivities, and many people use the same model. Imagine all books on earth were written by just 5 people – and worse, they could not remember what they had written yesterday. You would quickly get tired of reading the same flourishes and patterns over and over.” – hashmap
“I predict by the end of the year they'll have a style/voice selector for responses.” – seje
“The most difficult thing in doing it yourself is knowing what to ask for… asking for something more nuanced… requires the correct lingo.” – wongarsu

Theme 3 – Widespread AI writing threatens authentic human voice and expression

“Why would you support this sort of projects? LLM text should be spottable for as long as possible, otherwise human communication and expression on the internet will be killed.” – f311a
“I stopped using reddit entirely because it's bots talking to other bots on threads posted by bots… I no longer know if a human being is ever reading them.” – chownie
“Why would one leave writing to the clanker though, given that expression is literally the only thing worth doing as a human?” – hypfer


🚀 Project Ideas

CursorGrace: Subtle Collaborative Cursor Overlay

Summary

  • Provides a lightweight, configurable cursor-sharing overlay for blogs and article pages that reduces visual distraction while preserving awareness of concurrent readers.
  • Core value: lets publishers enable real‑time cursors without driving readers away, addressing the “distracting cursors” complaints on HN.

Details

Key Value
Target Audience Blog owners, news sites, educational platforms that want to show live reader activity
Core Feature Customizable cursor dots (size, color, opacity, fade‑out delay) with a global hide/show toggle and optional hover‑only mode
Tech Stack HTML5 Canvas / SVG overlay, lightweight JavaScript (Vanilla or React), WebSocket via Trystero for sync, optional Service Worker for caching
Difficulty Medium
Monetization Hobby

Notes

  • HN users complained “the very distracting cursors of others” and immediately hid them; a gentler, user‑controlled overlay would keep the feature without the annoyance (see anilgulecha, mullanba, markgavalda).
  • Enables discussion about awareness UI design and could be extended to show aggregate heatmaps, opening a practical utility for site analytics.

VoiceFit: Personal LLM Style Adapter Service

Summary

  • Allows writers to upload their own text samples and receive a LoRA‑style adapter or prompt template that makes any LLM output match their personal voice, reducing generic AI “slop”.
  • Core value: turns a generic model into a personal writing assistant that sounds like the user, addressing the desire for style/voice selectors and frustration with AI tics.

Details

Key Value
Target Audience Bloggers, professionals, students who use LLMs for drafting emails, reports, or creative work
Core Feature Upload writing samples → automated extraction of style cues (word choice, punctuation, sentence length) → generation of a reusable adapter or prompt snippet
Tech Stack Python backend (FastAPI), HuggingFace Transformers + PEFT for LoRA training, optional vector store for style embeddings, React frontend for upload/download
Difficulty High
Monetization Revenue-ready: Subscription tier ($9/mo for unlimited adapters)

Notes

  • HN commenters asked for a “style/voice selector for responses” and shared personal experiments feeding their own writing to Claude to match voice (jamesponddotco, techjamie). VoiceFit automates that workflow.
  • Could spark discussion on personalization vs. privacy, and offers practical utility for anyone wanting LLMs to sound less robotic and more authentic.

SlopSniper: AI‑Tic Highlighter & Rewriter

Summary

  • Highlights common AI‑generated writing tics (overused em dashes, repetitive phrases, overly formal diction) in any text and suggests human‑like rewrites.
  • Core value: gives writers an instant “de‑slop” check before publishing, helping them avoid the telltale signs that make readers dismiss content as AI‑generated.

Details

Key Value
Target Audience Writers, editors, marketers, and anyone who posts long‑form content online
Core Feature Real‑time detection of AI slop markers with inline suggestions; optional one‑click rewrite using a tuned de‑slop model
Tech Stack TypeScript (React) for editor UI, Rust/WASM for fast regex‑based detection, optional HuggingFace model (e.g., deslop‑bert) for rewrite, Electron or browser extension
Difficulty Medium
Monetization Hobby

Notes

  • Users noted that em dashes, certain phrasing, and generic patterns are “major indicator of AI writing” (literalAardvark, ahmedfromtunis) and asked for tools to reduce such tics; SlopSniper directly addresses that pain (see mrweasel, j45).
  • Provides a concrete tool for discussion about AI detection ethics and offers practical utility for improving readability and credibility of AI‑assisted writing.

Read Later