Technology

You Can Now Tell Facebook What You Want to See

Martin HollowayPublished 5w ago2 min readBased on 1 source
Reading level
You Can Now Tell Facebook What You Want to See
Photo by Lucas Pezeta on Pexels

Meta is letting Facebook users tell the Feed directly what they want to see. Engadget

The option sits in a new "your algorithm" menu on individual Feed posts and Reels. You open it from a specific post or Reel. Settings you change there shape what you are shown next.

The menu offers three inputs in one place. You can add topics to an interests list. You can add topics to a "not interested in" list. You can also type in your own words what you want more of in Feed.

Meta says this is possible because its ranking systems now use large language models, a type of AI that understands everyday language. That matters for how the inputs work. Typed requests and topic choices must be understood for meaning before they can guide ranking. A ranker built this way can use that kind of plain, user-written input directly.

The company brought a similar feature to Instagram first. The Facebook rollout extends the same idea of listed likes, listed dislikes, and written descriptions to Feed and Reels.

In my view, the interesting change is from guessing to asking. Feed ranking has long learned from behavior like watches, likes, time spent, and skips. Those clues are plentiful but messy. They mix what you saw with what you liked. It is like having a shopkeeper guess from what you glanced at, rather than telling them what you want. Lists you fill in yourself give clearer answers. They offer less data but more exact data.

Worth flagging is how typing changes the setup. A sentence can express combinations and exceptions that checkboxes handle badly. "More trail running, less marathon training content" is simple to write and hard to fit into preset categories. For this kind of ranker, that sentence can be used directly when picking posts. The tradeoff is vagueness. People write unclear requests, change their minds, and leave old entries in place. A working system must sort out clashes between the typed request and the lists, and between new requests and old history.

Another practical issue for Feed builders is how broad a preference should be. Starting from a single post makes sense. Annoyance or delight happens at the post. But it is unclear if that choice means a whole topic, a format, a creator, or just a passing mood. That affects what gets picked from the pool of possible posts. Clear controls mean less guessing, but they ask more upkeep from users. Most people will not tend lists carefully. The systems that gain most will treat a short statement as a strong starting hint, then keep learning from behavior rather than sticking only to the exact words.

The broader context here is user control as a product feature. Giving people readable controls they can edit does not replace ranking work. It adds a useful new input alongside the existing steps that fetch posts and predict taps. When it works, the payoff is faster agreement on what relevant means for you, with fewer wasted views. That is a practical gain for daily use, and a useful path to keep exploring.