Will AI replace software engineers?

Anthropic’s 'harmlessness training' is an interesting approach to alignment.

We need more debate on interpretability vs. pure performance.

I think we're still scratching the surface of true AI alignment.