Do Social Media Bans Work? + A Conversation About A.I. Consciousness + Tool Time

What It Covers

Kevin Roose and Casey Newton examine teen social media bans across eight countries, analyzing early Australian data showing 85% of teens still online after 90 days, then NYU professor Jeff Sebo discusses Anthropic's JSpace interpretability research and the empirical framework for studying AI consciousness, followed by tool demonstrations including Glaze vibe-coding app and Gemini Spark monitoring.

Key Questions Answered

• •Social Media Ban Enforcement: Australia's law requires platforms take "reasonable steps" to remove minors, not mandate specific methods. Platforms currently use age inference signals — messaging patterns, linguistic markers — which take months to calibrate. Eighty-five percent of Australian teens remained on social media 90 days post-ban, prompting Australia to introduce legislation doubling fines against non-compliant platforms. Enforcement effectiveness scales gradually, not immediately.
• •US Age Verification Legal Landscape: Texas's App Store Accountability Act survived Supreme Court review after the Fifth Circuit overturned a lower court block. The law requires minors to link accounts to a parent or guardian who must approve every app download. Four additional states passed similar laws, and California's less restrictive version — requiring age collection at device setup — takes effect January 1. Pew research shows 60% of US adults support a ban for under-16s.
• •Direct Harm vs. Population-Level Research: Jonathan Haidt's framework shifts the debate from population-level mental health effect sizes — where studies show small or no effects — toward direct, documented harms: millions of annual reports of grooming, sextortion, and scamming targeting minors. This reframe argues that removing 13- and 14-year-olds from platforms would spare millions from specific, identifiable harms regardless of broader mental health correlation data.
• •AI Consciousness Empirical Framework: Jeff Sebo's new report from the Center for Mind Ethics and Policy argues that determining AI consciousness requires synthesizing three evidence types simultaneously: behavioral evidence (how models act), internal evidence (how models process information), and developmental evidence (how training shaped them). Relying on any single evidence type produces overconfident conclusions in either direction, mirroring how animal consciousness research integrates behavior, neuroscience, and evolutionary history.
• •JSpace Global Workspace Significance: Anthropic's interpretability research identified an internal processing space in Claude — named JSpace after the Jacobian mathematical concept — where privileged representations are staged before output generation. When this workspace is disabled, Claude loses advanced reasoning capabilities. The finding parallels global workspace theory, a leading neuroscience framework for consciousness, though Anthropic explicitly states it does not confirm phenomenal consciousness or sentience in current models.
• •Vibe-Coding Desktop Apps with Glaze: Glaze, made by Raycast, lets non-technical users build Mac desktop apps visually in under two hours without scaffolding setup. Users describe changes by circling interface elements directly. Casey Newton built a searchable archive of all Platformer articles using an Anthropic API key and a separate Nightwing-themed to-do list app with AI-generated task illustrations. Glaze offers a free tier with limited credits; full access costs $20 per month.