Why does the system still need a reranking stage after the ranking model has already scored everything?
A ranking model optimizes for historical engagement patterns, which can create a feedback loop — repeatedly surfacing content from whatever the user clicked last week, even if that produces a repetitive, fatiguing experience. Reranking applies product-level judgment on top of the model's raw output: enforcing diversity across subreddits, boosting subscribed content, and adjusting emphasis based on how a specific user tends to engage — corrections a model trained purely on past clicks won't make on its own.
Answered in
Reddit's Hardest Notification Problem Isn't Ranking, It's RestraintReddit's notification pipeline spends as much effort deciding when to stay silent as deciding what to send — budgeting runs before ranking.
Read the full analysisOther questions this article answers
- Why does Reddit's notification system decide a daily budget before it even looks at candidate posts?
- Why use causal modeling instead of simple correlation to set notification budgets?
- What problem does the two-tower retrieval architecture solve that simple subreddit-subscription filtering doesn't?
- Why does the ranking model predict multiple outcomes (click, upvote, comment) instead of just one?
More how ai actually works questions
- How many new researchers can receive the next OpenAI grant?
- When do applications open?
- Does the classifier make a final decision about an image?
- Can any developer call the classifier?
- How can Cursor answer questions about an entire codebase without storing the code?
- If embeddings are stored instead of source code, can the original code be reconstructed from them?
- Why does Cursor use a Merkle tree to keep the index synchronized instead of just re-scanning changed files?
- What happens when the server needs the actual source code, not just its embedding?
Every answer on Crashtech is written by the editor of the article it comes from — never auto-summarised. Browse all answers or the How AI Actually Works beat.