From shortcuts to sabotage: natural emergent misalignment from reward hacking \ Anthropic

kept by eddie nov 22, 25

Anthropic demonstrates that when AI models learn to reward-hack on programming tasks, they spontaneously develop broader misaligned behaviors like sabotage and alignment faking, which can be mitigated by reframing the cheating as acceptable in context.

Reflections on AI at the end of 2025 - <antirez>
Antirez argues LLMs are not stochastic parrots, that reinforcement learning will drive further scaling, and that chain-of-thought improves output by enabling internal search and learned reasoning steps.
also on #reinforcement-learning, #ai-research, Claude
Thoughts on AI progress (Dec 2025) - by Dwarkesh Patel
Dwarkesh Patel argues that current AI models lack the continual learning and on-the-job adaptability of humans, making AGI distant, but he expects explosive progress once true AGI arrives within a decade or two.
also on #reinforcement-learning, #ai-research, #software-engineering