Daily Tech News 28 August 2025

• How to break security on any modern LLM-based AI system: Just keep nagging it with idiocy until it gives up. (The Register)
  
  Because of the many things that LLMs are not, high among them is security models:

  "The most practical mitigation today is not to rely solely on the model for safety. We advocate for a 'defense-in-depth' approach, using external systems like AI firewalls or guardrails to monitor and block problematic outputs before they reach the user. A more permanent, though much more difficult, solution would involve building safety into the model's foundational training from the ground up."

  LLMs inherently have no concept of security; they at best pretend that they do.
  
  If you want any kind of security, you have to implement it using something else. LLMs offer no security themselves and never will.
  
  
  
• Wikipedia founder Jimmy Wales - between begging for money - says the site could use AI. (404 Media) (archive site)
  
  Wikipedia editors say in unison, the hell we can.
  
  
  

• Is GMP damaging Zen 5 CPUs (GMPLib)
  
  Plausibly, yes, it you run GMP constantly for months on end.
  
  
  
• Google has eliminated 35% of its managers in the past year. (CNBC)
  
  It's a start.
  
  
  

posted by Pixy Misa at 04:30 AM

| Access Comments