Oops I shouldn’t have pointed to context window limitations. Probably the real reason the AI is incentivized to compress its chain of thought is because of its output token budget/effort level, not context window limitations (they can remember a million tokens and compact long chains of thought).
Oops I shouldn’t have pointed to context window limitations. Probably the real reason the AI is incentivized to compress its chain of thought is because of its output token budget/effort level, not context window limitations (they can remember a million tokens and compact long chains of thought).