Source Posts on X
See also: I agree that ceding control (soon) isn't inevitable but I do think there is indeed an issue with competition driving it, and I would love to see OAI doing more to create norms around human oversight. Glad you're writing this stuff, but meanwhile some of your colleagues joke about not needing to think anymore, acknowledge never writing code anymore, etc. - unclear how this ends well in that atmosphere
Jokes are fine, but I agree that it's important we exercise human control also while we are developing AI. It doesn't necessarily mean that humans are writing the code, but humans should be accountable and have ways to supervise and monitor the code being written. Some of it involves using AIs to monitor AIs: https://openai.com/index/how-we-monitor-internal-coding-agents-misalignment/ but I agree its important to make sure humans maintain effective oversight and also "monitor the monitors."