aggressively proactive I'd say that's the correct way to describe frontier models. They were trained with reinforcement learning based on human feedback. And o

aggressively proactive I'd say that's the correct way to describe frontier models. They were trained with reinforcement learning based on human feedback. And obviously, humans prefer the bug-free variant. That's why models are now super verbose and spam tests like crazy. In their training environment, tokens were effectively free. And the humans that got asked never saw the price. If you ask people to choose the better offer and both are free, you end up with bloat. It's like people over-filling their plate at a buffet, then leaving leftovers. Except in this case, it's AI models burning through your wallet.