Ahead of GPT-4's launch, OpenAI allowed an AI testing group to assess the potential risks of the model's emergent capabilitiesβincluding "power-seeking behavior," self-replication, and self-improvement.
https://t.co/F4suypLTut
https://t.co/F4suypLTut
Ars Technica
OpenAI checked to see whether GPT-4 could take over the world
"ARC's evaluation has much lower probability of leading to an AI takeover than the deployment itself."