A user says Anthropic confirmed its models were distilled
Stolen Thoughts claims encrypted reasoning blocks from Anthropic, OpenAI and Google APIs are interchangeable across sessions, users and models—and says it uses this to decode hidden reasoning at scale.
TLDR
A user sharing Stolen Thoughts said Anthropic had confirmed that its models were distilled in the way the site describes. The site claims encrypted chain-of-thought blocks returned by Anthropic, OpenAI and Google APIs are interchangeable across sessions, users and models. It says it exploits that interchangeability to decode hidden reasoning at scale.
Combined views
300K
5 Sources, first seen 20d ago