AI’s most closely-guarded secret, besides the frontier model itself, is its ‘chain of thought’. Stealing it lets rivals copy these models and potentially match them.

Unsurprisingly, closed AI labs like OpenAI and Anthropic hide them from the world. But because of a fatal error, these hidden thoughts were never truly hidden and were open for the world to see.

Not only is the mistake deeply embarrassing, but worse, even after patching, it might not matter either way. — Read More

#chatbots