Feed Curation

Definition

Feed curation is the deliberate selection of recurring people, channels, formats and interactions that shape one’s information environment. It includes both choosing what to receive and deciding what behavioral feedback to send to a recommendation system; it is not a promise to escape mediation entirely.

Current Synthesis

The practical problem is not merely the volume of information. Ranking systems preselect what can be seen, learn from attention and present partial views as relevant; users then adapt their habits to those views. This carries Attention Industrialization’s critique of mass-produced mental intake into a choice of daily inputs and feedback. Curation offers several different interventions: reduce low-value recurring inputs, avoid reinforcing content one does not want recommended, compare dissimilar filters, and reserve time for complete arguments and unfamiliar perspectives. The examples range from one person’s strict input diet to a researcher’s more qualified account of recommendation effects. They support a discipline of deliberate choice, not a quantified guarantee that fewer contacts or more opposing views will improve everyone.

Key Claims

• Selective reduction of contacts and subscriptions can create room for thought and closer attention to chosen relationships, as an individual practice rather than a universal threshold.
• Clicks, watch time and explicit preference controls both inform ranking and shape the profile through which future material is presented to a user.
• Refusing reflexive engagement with anger bait or entertainment drift can be a curation choice, because commenting and lingering may train the very feedback loop one hopes to avoid.
• Comparing platforms and slower media can expose the limits of any single filter, while the “cave” image remains a philosophical analogy rather than an algorithm audit.
• Choosing long-form material and selectively testing one’s worldview with different perspectives can counter personally felt phone and AI passivity, without presuming disagreement always yields understanding.

Evidence

• Reducing recurrent inputs: E45 孟岩对话李继刚：人何以自处 recounts Li Jigang / 李继刚’s pruning of thousands of WeChat contacts to a smaller circle and a few people’s posts, limiting public accounts and RSS subscriptions while retaining paper books and AI-assisted reading of papers. Meng Yan / 孟岩’s “your feed is your fate” frames how repeated inputs might form attention, memory and desire; the episode treats lower-level constraints as room for higher-level thought and relationships, not a measured effect. Its AI-assisted-paper workflow also illustrates a risk: faster synthesis can amplify the consequences of an unexamined input diet.
• Participatory ranking: 159.算法的六副面孔：它是如何从处理数据，变成定义我们是谁的 describes inclusion, prediction, ranking and calculated user profiles, including likes, skips and “not interested” signals. This gives a mechanism for curation beyond the follow list.
• Feedback-aware restraint: 164.算法的“兔子洞”：为什么你总在看完新闻后滑向娱乐？｜对谈黄圣淳教授 recommends noticing one’s feedback before reacting to rage bait and reports a source-scoped YouTube audit of news-to-entertainment recommendation chains, an example of Algorithmic Entertainment Redirect / 算法娱乐重定向.
• Multiple lenses: 159.算法的六副面孔：它是如何从处理数据，变成定义我们是谁的 argues that mixed platform use can reveal divergent filters of public relevance; 167.柏拉图、卢梭、哈耶克、阿伦特四大哲学家会如何解释算法时代？｜串台独树不成林 uses Plato’s cave and reason-outsourcing analogies to explain why context and comparison matter. Both are episodes of the same algorithm series, not independent platform tests.
• Long-form and worldview testing: 154.四十岁感言：不做那只温水里的青蛙 gives 大卫翁 / David Weng’s personal account of heavy phone use, AI answers, books and podcasts as slower alternatives, and intentional attention to disagreeable views. It shares a show and speaker with the algorithm-series notes.

Counterevidence & Qualifications

Episode 164 says evidence for universal, sealed information cocoons is weak: incidental exposure exists, and contact with opposing views can increase affective polarization. Its YouTube finding concerns a particular audit design, not every feed. The philosophical treatment in episode 167 is not empirical validation of what an individual recommender does. Li’s input restrictions and Weng’s long-form habits are personal testimony, not causal studies of attention or identity; multiple episodes from Qizhulou Yan Binke should not be counted as independent replications. Some filtering is unavoidable and useful under information overload, so the aim is inspectable choice rather than an unfiltered world.

What Changed

• Curation now includes outgoing behavioral signals as well as incoming subscriptions.
• Reducing volume and increasing diversity are treated as complementary but distinct choices.
• Claims about filter bubbles are narrowed by incidental exposure and polarization evidence.
• The cave metaphor is separated from platform-specific research and personal testimony.

Related Concepts

• Platform Feedback Loop / 平台反馈循环 - explains how curation of reactions can alter later recommendations.
• Algorithmic Media Literacy / 算法媒介素养 - supplies the reflective skill of noticing ranking and one’s own feedback.
• Algorithmic Diversity Dividend / 算法多样性红利 - describes the potential benefit of comparing non-identical filters.
• Autonomy Under Information Flow / 信息流中的自主性 - names the agency problem that motivates deliberate input choice.
• Personal Knowledge Ecology - connects selected inputs to later questions, notes and judgments.
• Public Relevance Algorithms / 公共相关性的算法 - describes the ranking systems that decide which material can become visible or apparently relevant.
• AI Use Pacing - addresses the tempo of AI-mediated input, alongside selection of sources.
• Human Agency Under AI - situates feed choice within the wider question of delegating judgment to AI.
• Wet-State Human Agency - retains bodily, emotional and relational volition beyond an optimized information diet.
• AI-Assisted Reading - is one way to process selected papers without making the model the sole source of judgment.
• Flow Environment Design - complements input selection by shaping conditions for sustained attention.
• Rumination Vs Reflection - distinguishes unproductive replay of noisy inputs from deliberate thinking.