social media is about to lose its only advantage

The one thing Twitter did well was showing you stuff you wouldn't have found on your own. You followed some people, the algorithm noticed patterns, and eventually your feed became this weirdly specific stream of exactly the kind of content you cared about. That was the product. Not the tweets themselves, not the reply guys, not the trending topics. The curation.

That advantage is about to disappear.

An LLM can read thousands of blog posts, newsletters, and articles in seconds. It doesn't need to track your clicks or sell your attention to advertisers to figure out what you care about. You can just tell it. "I'm interested in distributed systems, agent frameworks, and weird history facts. Less crypto stuff please." Done. Better curation than Twitter ever managed, and you didn't have to like or dislike a single thing to get there.

the like button was always doing two jobs

Think about what happens when you like a post. Two completely different things happen at once. First, you're telling an algorithm "show me more of this." Second, you're telling other people "I found this worth reading." Filtering signal and social commentary, packed into one click.

That bundling made sense when the algorithm needed your clicks to learn what you wanted. Likes and dislikes were the training data. The platform watched what you engaged with and tuned your feed accordingly. Your social expression was the price of admission for decent filtering.

LLMs don't need that signal. You can just describe what you want in plain language. "More long-form analysis, fewer hot takes. I like when people cite primary sources. Stop showing me engagement bait." The model adjusts. No clicking, no training loop, no behavioral surveillance. Just a conversation about what you want to read.

So the like button loses its filtering job. What's left?

liking becomes commentary

The part that remains is the social signal. Not "optimize my feed," but "here's what I think." With filtering handled separately, the like transforms into something closer to sharing, or a reaction. What does Paul Graham think about this new programming language? What papers is Karpathy reading this week? That's interesting information, and it has nothing to do with ranking your personal feed.

People want to know what interesting people think about things. That desire doesn't go away just because an LLM handles your filtering. If anything it gets more important. When your filtering is personal and private, the social layer becomes the only way to discover things outside your own bubble. Someone you respect reacts to a post about urban planning and suddenly you're learning about zoning laws. That serendipity is valuable, but it's a commentary function, not a filtering function.

The point is these two things don't need to be jammed together anymore. We want LLMs to filter. That's one system. And separately, we want a way to broadcast what we think about things we read. That's a different system. Conflating them was an accident of technological limitations, not a design choice worth preserving.

what this looks like in practice

For the filtering: needle. You add your RSS feeds, subreddits, keyword searches, whatever sources you read. An LLM reads everything, scores it against your interests, and gives you a ranked feed. You open it in the morning and there's a brief, a few paragraphs covering what happened across your interests overnight. Factual stuff, trends, who said what. You give feedback in plain language, "more of this because..." or "less of this, too surface-level," and your interests evolve naturally. No likes, no dislikes, no clicking little arrows. Just tell it what you want.

For the commentary: likefeed. You publish your reactions as a public feed. Anyone can subscribe the same way they subscribe to a blog's RSS. No platform owns it. No algorithm uses it to decide what you see next. It's just "here's what this person found interesting," available to anyone who cares. Subscribe to a few people whose taste you trust and their reactions flow into your reader as another source for your LLM to rank.

Two systems, cleanly separated. Filtering that works for you, not for advertisers. Commentary that's honest, not a side effect of training an algorithm.

Social media bundled them together because it had to. It doesn't have to anymore.