PhysicsForums and the Dead Internet Theory

Original Article ↗ Hacker News Discussion ↗

Account Hijacking, Backdating, and Trust

  • Many see reusing dormant forum accounts and backdating LLM posts as a serious breach of the implicit social contract of online communities.
  • Concerns include impersonation, potential defamation, erosion of individual professional reputations, and loss of trust in archives people still use for learning.
  • Some find the site owner’s explanation (“internal test” to answer long-unanswered threads) unconvincing and “shady,” especially combined with SEO motives.

Technical and Legal Countermeasures

  • Ideas floated: PKI and PGP-signed posts, web-of-trust schemes, proof-of-work to make posting expensive, cryptographic markers linking posts to independent archives, and “content under your control” protocols (e.g., ATProto-like).
  • More advanced proposals: zero-knowledge proofs, verifiable credentials (OpenID4VCI), e-passport-based proofs-of-humanity.
  • There is debate over practicality, usability for non-technical users, and risk of creating Orwellian identity systems.
  • Legal angles mentioned include trademarks and copyright, but commenters doubt enforcement will keep up with LLM rephrasing.

Identity, Verification, and Privacy

  • Suggestions range from invite-only and real-ID-gated communities to anonymous-but-verified “human” credentials.
  • Tension: stronger identity systems might fight bots but destroy remaining online privacy and do not prevent humans from posting AI slop.
  • Some point to socialized trust (follower graphs, reputation) as more workable than centralized ID.

Attitudes Toward LLM Content

  • Strong sentiment that people don’t want AI-generated answers in human discussion spaces, especially when undisclosed or used for emotional/mental health support.
  • Users describe active strategies to detect and avoid “ChatGPTese,” including date filters and style tells, and wish for a “no generative content” search filter.
  • Others note misclassification risk: cautious, formal human writing can look AI-like.
  • AI images are seen as less problematic in some creative contexts but disliked when used as low-effort filler (e.g., kids’ books, blog headers).

Decline of Classic Forums and Search Changes

  • Several trace the decline of forums/blogs to Google’s ranking changes (Panda, removal of “Discussions/Blogs” filters), which reduced visibility of niche communities.
  • Old forums are remembered for deep, continuous threads and camaraderie; modern spaces (Reddit, Discord) are seen as more transient, cliquish, or spam-filled, with lore lost inside closed chats.

Dead Internet Theory and Scope

  • Some treat PhysicsForums’ heavy AI contamination as emblematic of a broader “dead internet” trend; others argue this is confirmation bias and that large vibrant areas still exist.
  • There is concern that AI-generated content will poison future AI training and that similar dynamics are emerging in hiring (AI vs AI in resumes and screening).