If you've been online in the past few years, you've probably gotten used to the flood of spam. LLMs have made it even easier — generating ads, fake reviews, and phishing links that look like human-written text at virtually zero cost.

Reddit recently published a blog post saying they've developed a new tool using LLMs to cut down spam. Ironically, most of that spam is also created by LLMs.

The platform has no choice but to fight fire with fire.


*Image credit: TechCrunch*

According to Reddit, this new system blocks 23 million spam views per day and identifies and stops about 25,000 new spam posts and comments daily.

Social platforms have been doing spam filtering for a long time, but Reddit says this time the results are better than before.

Their blog post states:

> "We use LLMs to capture highly covert, coordinated fake behavior patterns and artificial hype that our old systems couldn't detect."

Specifically, from January to March this year, the rate at which users encountered spam decreased by 20% compared to the previous three months.

YouTube, Meta, and Instagram allow users to post AI-generated content as long as it's clearly labeled. TikTok even lets users adjust how much AI content they want to see.

If platforms can detect AI content faster, in theory they can also flag hate speech and other violations more quickly. However, platform governance experts have been reminding us that AI moderation must be paired with human review for the best results.

Reddit's approach this time sets an example for all platforms plagued by LLM-generated spam. Using magic to defeat magic sounds absurd, but it might be the most effective solution for now.