Some Raw thoughts on @OpenAI Operator - 0. Can Operator pass a captcha test? 1. The take control option is amazing, additionally. Specifically I love that we can go back and see what steps the Agent took to reach the current state. But why does it need us to tell it what changes we made ? It should be able to figure this out on its own. 2. The fact that this does not need any API access and interacts with the websites visually is great. This is a baby step towards AGI.. and eventually physical AI 3. How comfortable will people be when entering their card information on a Remote browser? This can be a big barrier in adoption 4. While the operator aims to mimic a human behaviour, does it raise flags when accessing websites at a quicker than human pace (which it should be able to in a few years) 5. Can Operator learn and adapt to user preferences over time, like my favourite restaurant, shopping stores etc? 6. Does Operator recognize regional differences? For instance, would it seamlessly navigate a website with a different languages outside English? 7. Could it detect potential scams or phishing websites before proceeding, or entering sensitive information, especially on less familiar domains? 8. How does Operator decide when to ask for user confirmation and when to act autonomously? What happens if it accidentally remembers the previously entered card information used the last time and enters it without confirmation?