AI versus AI: How AI helps us to evaluate contents and stop generated nonsense
How do we prepare in Seznam for the landslide of generated content? Artificial intelligence is already helping our editors, moderators, and fact-checkers make sure that the content on our platforms is trustworthy.
We'll show you what works for us and where we see AI's limits:
- AI tools for moderating and fact-checking discussions: Our editors are using LLM to debunk manipulative articles and comments. They are also using KInIT's fact-check database to do semantic searches for independent third-party fact-checks. The artificial intelligence in our discussion platform automatically approves benign discussion posts and scans about 80% of comments (its work is randomly checked by our editors). We are currently deploying an AI tool for our discussion management that automatically assigns fact-checks to manipulative discussion posts.
- Our editors at Seznam Medium and soon also at Newsfeed use our own AI-generated content detector: It's trained on Seznam Zprávy and Novinky.cz articles and runs on our own BROWCzech language model. For long texts (in this case articles on Seznam Medium), it can determine with high precision which parts of the text were generated and which were written by a human. This helps editors to detect generated articles and prevent Seznam Medium to be a dumping ground for generated texts. We don't mind when authors and journalists help themselves to AI tools, but it's important to us that they disclose it, source them correctly, and that their texts pass through the human editorial review and editing process.
- AI helps detect clickbait articles in the Newsfeed: We've been using an AI classifier for clickbait headlines since 2022. The "clickbaitiness" of an article is one of signals for evaluating the quality of articles in the feed. Clickbait articles are penalized by our recommending algorithm and we try to demote them in the personalized feed so that Newsfeed does not become a glut of clickbait, but offers readers as much quality and trustworthy content as we possibly can.
Peter Jančárik
Peter coordinates Seznam's projects in evaluating media quality and in fighting against disinformation. He has worked in corporate communications and brand safety for more than 15 years. In addition, he had co-founded Konšpirátori.sk, a project which lists fake news websites and helps advertisers to blacklist them in their ad campaigns.