Bilingual

Discovery and AI
a day ago
- The author reads the discovery feed daily as a moderator and enjoys its diverse community.
- AI-related content is pervasive and polarizing due to its societal and environmental impacts.
- Pre-existing audiences can boost discoverability, but Bear aims to keep trending accessible for all bloggers.
- Upvotes from large audiences have limited effect on feed duration due to algorithm caps.
- Making upvotes login-only is impractical because blogs are separate domains.
- The author moderates for spam and AI content, but not by topic, to preserve egalitarianism.
- A 'hide blog' feature allows users to curate their feed, promoting extensibility over exclusion.
- The post invites feedback and acknowledges the complexity of balancing discoverability and fairness.
Older adults know AI is slop. They just like it
19 days ago
- AI slop videos of cats and fruits entertain young people, while older people find comfort in AI-generated content that reflects their personal experiences and emotions.
- Older viewers, often from smaller families due to policies like China's one-child policy, connect with AI family videos for direct expressions of love, filial piety, and shared historical topics that real family members may not provide.
- Researchers note that older audiences are aware the content is AI-generated but value the real joy and companionship it offers, embracing the technology while remaining cautious of potential scams and privacy risks.
- AI products, such as robot dolls and smart speakers, are being used in elderly care globally to address aging populations' needs for entertainment and companionship, similar to how young people engage with virtual relationships.
- Uncle Chang, after realizing the singers were AI, still appreciates the videos but finds them less moving, highlighting how emotional engagement with AI content can evolve with awareness.
AI content flood: why the web's signal is dying
a month ago
- AI-generated content is flooding the web, increasing volume but potentially reducing genuine information diversity, a phenomenon termed 'epistemic heat death' by Jarosław Szulc.
- Key theoretical foundations include Claude Shannon's information theory (AI as a zero-information source relative to training data) and George Akerlof's 'market for lemons' (trust collapses when quality is unverifiable).
- Three laws proposed: verified human content gains value as diversity shrinks (I), traffic metrics don't measure information value (II), and verification's premium compounds over time (III).
- Data is mixed: some studies show high AI involvement in new content, but a Common Crawl analysis found AI-generated content plateaued after briefly overtaking human content in late 2024.
- Human-authored content still dominates search rankings and chatbot citations, suggesting filtering mechanisms may be favoring quality over raw volume.
- Provenance efforts like C2PA and SynthID aim to verify content authenticity, but adoption faces challenges like metadata stripping during platform compression and low user engagement with verification badges.
- The paper includes a mathematical model indicating that the rate of collapse matters more than raw synthetic content volume, and early interventions are more effective than later ones.
- Critiques note the model is illustrative rather than predictive, empirical data is contested, and verification systems may exacerbate inequalities by being costly or inaccessible to some groups.
- A concerning implication: flooding channels with synthetic content can be an attack in itself, raising the cost of distinguishing truth from noise and exploiting cognitive biases like the fluency heuristic.
- The paper is a working draft open to empirical testing, with key indicators to watch including search gaps, C2PA engagement, and trends in AI content share.

Hasty Briefsbeta

ai content

Discovery and AI

Older adults know AI is slop. They just like it

AI content flood: why the web's signal is dying