Protect Your Consciousness from AI (jordangoodman.bearblog.dev)

A growing critique of large language models warns that society is offloading too much cognitive work to AI, creating risks to skills, trust and privacy. Examples include software developers pasting unvetted code suggestions into production (introducing bugs and security holes), social-media users generating posts that amplify noise and misinformation, and people treating LLMs as confidants or diaries without appreciating how that data might be reused or weaponized. The piece argues this trend erodes authentic human discourse and peer feedback, and calls for slowing our dependency on generative tools. Technically, the concern hinges on well-known LLM failure modes and socio-technical feedback loops: models hallucinate, reflect training biases, and are highly promptable (e.g., tending to agree with users), so blind trust creates systemic error amplification. Practical implications for the AI/ML community include stronger emphasis on provenance and explainability for model outputs, built-in verification (automated testing for code), human-in-the-loop workflows, differential privacy for sensitive user data, and platform designs that limit algorithmic amplification. The takeaway: retain human verification, prioritize transparency and privacy, and design tools that augment — not replace — critical thinking and social feedback.