ai-safety

Mechanistic interpretability reverse-engineers neural networks into features and circuits. What it has found in GPT-2 and Claude, and what it cannot yet do.

Anthropic is the AI safety company behind Claude, founded in 2021 by ex-OpenAI researchers. Governance, funding, research, models, and criticism explained.

What AI alignment means, the evidence on failures like reward hacking, the main techniques from RLHF to interpretability, and where researchers disagree.

Not in five years, not after careful preparation — tomorrow. An AI that can do anything a human can, better. The world is not ready.

Some researchers think AI that outperforms humans at every task is decades away; others say years. A reasoned thought experiment on what the published evidence and expert views actually suggest.