KI-Chatbots Destroying the Internet (gnulinux.ch)

A trenchant critique argues that general-purpose LLM chatbots are degrading the internet’s information ecosystem and creating an unsustainable economic and legal apparatus. Economically, massive capital spending by a few giants (Deutsche Bank/Goldman cite roughly $368B in cumulative AI capex) is propping up short‑term growth—much of it funneled into Nvidia‑optimized hardware and reciprocal investments—creating a concentration risk that looks more like a self‑funded bubble than productive adoption. Legally, models have been trained at scale on scraped web content and legacy media; a few licensing deals exist but many creators can’t afford litigation, exposing the industry to copyright and criminal‑receipt (Hehlerei) dilemmas and an uncertain business model if data must be paid for. Technically and socially, the piece warns of a feedback loop: LLMs increasingly replace index search and draw heavily on internet content (including Wikipedia), then generate content that becomes new training material—an “inbreeding” that impoverishes signal quality. Independent audits show systemic failure modes: a European Broadcasting Union study found roughly 45% error rates across major assistants, and students report heavy dependence on chatbots with potential harms to learning and creativity. Employment impacts are mixed—Meta cut ~600 AI roles while surveys (Bitkom) show most companies expect little net job growth—but the bigger risk is collapsing public knowledge infrastructure and rising misinformation. The author concedes that domain‑specific ML (e.g., medical imaging) remains valuable, but urges urgent reevaluation of training data practices, economics, and regulatory oversight.