Distilling Frontier AI Models
marginpoints.com> Obfuscation and deception might be a big part of how countermeasures are implemented.
That Anthropic already did this destroyed any trust I had left in them.
Distillation is not "stealing" in my mind. It is learning from each other and now that we have open weight models in the same class, we won't need Anthropic and OpenAI to distill into smaller or newer models.