OpenAI say "the reasoning system selected which answers to submit with no bespoke test-time harness whatsoever" – but were the answers themselves developed with the assistance of tools?

→ View original post on X — @simonw,