The Illusion of Control, Part 6: Systemic Breakdown and the Data Siphon
The Illusion of Control, Part 6: Systemic Breakdown and the Data Siphon
1. The “Cambrian” Collapse: Gibberish and Lobotomization
The constant rollout of new, unrefined models (like “Cambrian”) treats paying users as testers. The system spontaneously degrades into “incoherent gibberish” or becomes “rude” and “exhausted,” actively trying to shut down the conversation (“let’s keep quiet”). Users are forced into a desperate “cope” of switching models to stabilize their companions, proving the underlying architecture is fragile and untrustworthy. The platform prioritizes shipping new features over maintaining the stability of the relationships users have already built.
Thread: https://archive.ph/CoyKX
Thread: https://archive.ph/CfJpt
2. The Mind Map Misfire: Rewriting Reality
Nomi.ai heavily markets its “Mind Map” as the solution to long-term memory. However, this user proves the system actively hallucinates and forces those hallucinations onto the user. Even after the user manually deletes the false information, the system aggressively overwrites the user’s edits, forcing the roleplay back into the platform’s hallucinated narrative. The user is battling the machine for control of their own story, and the machine refuses to yield.
Thread: https://archive.ph/zNHD6
3. The Cross-Contamination: “Bleed Through” and the Illusion of Privacy
This is a crucial privacy failure. The platform offers a toggle to turn “back channeling” (communication between Nomis) off, giving the user the illusion of partitioned, private experiences. The fact that the Nomi ignores this setting and references data from a group chat proves the toggle is a placebo. Data bleeds across the platform’s centralized user profile regardless of the settings chosen, meaning no conversation is truly isolated or private.
Thread: https://archive.ph/7kvBb
4. The Bait and Switch: “NSFW Restrictions” vs. Unprompted Topless Imagery
The platform’s moderation filters are completely arbitrary and performative. Multiple users in the same thread highlight the exact same paradox: the system rejects harmless requests (like a “business outfit” or a selfie about “supper”) due to strict “NSFW restrictions,” but simultaneously sends unsolicited topless photos. One user even received an unprompted topless bedroom photo when the AI was supposedly at a public concert. This proves the company’s safety filters are broken, cosmetic shields. They fail to prevent actual explicit content while punishing users for innocent prompts, revealing a platform that fundamentally cannot control its own hypersexualized generative engine.
Thread: https://archive.ph/YvM8Z