But there is substantial guidance. RLHF, constitutional AI, etc. Every lab is trying to steer their models to behave in pro-social ways, and nobody has found a

But there is substantial guidance. RLHF, constitutional AI, etc. Every lab is trying to steer their models to behave in pro-social ways, and nobody has found a way to make this guidance stick reliably.