Gemma and LoRA power an iPhone app that filters AI slop locally
TurbulentTaro9 · reddit · 2026-07-23
The post argues that simple zero-shot prompting is too inaccurate for detecting AI-generated text or images, citing prior benchmarks that found prompt-only methods are barely better than random for smaller models.
Main point
Reliable detection needs models trained specifically for the task, not a generic LLM asked to judge content after the fact.
What they built
- Dedicated AI detection models for text and images
- A local-compute workflow using Gemma + LoRA
- A product called Bouncer with a “Remove AI Slop” button
The linked article explains that the tool is designed to filter AI-generated content locally on iPhone, and the post positions it as a practical product rather than a prompt trick.
More from Apps
- Google Docs and Drive add native Markdown support for human-AI collaboration — thisiskp_ · 2026-10-27
- Instinct launches Link balance to let agents spend within set budgets — jeff_weinstein · 2026-10-27
- Open Dungeon Master: open-source AI Dungeon Master app hits Play Store, runs 5e on a 16GB GPU — lebbi · 2026-10-27
- Show HN: Wraps, open-source Resend-style email API running on your own AWS SES — stewartjarod · 2026-10-27
- Sentry CEO builds AI-generated entity pages for whisky brands in side project Peated — zeeg · 2026-10-27
- Grok Impressions: Delegating Forms, Shopping and Payments to Grok Bot — XFreeze · 2026-10-27