- SliceOfAI
- Posts
- OpenAI Operator ๐ค ๐ฑ : An Agent browsing and doing task on your behalf ๐
OpenAI Operator ๐ค ๐ฑ : An Agent browsing and doing task on your behalf ๐
๐ค Some takes on: 1๏ธโฃ Navigating Captcha, 2๏ธโฃ Authentication for agents and, 3๏ธโฃ Will agents have biases and it's societal implication!
Yesterday, OpenAI released Operator, an interactive agent that can do tasks for you autonomously by accessing web browsers and do tasks on your behalf.
It's here everyone, now we don't have to spend time searching products on Amazon ourselves, we can just tell the Operator! Or can it? ๐
Disclaimer: I did not try the Operator, because I am not from US and obviously because it is only available in the $200 Pro subscription ๐
But I can still have an opinion on it right? ๐ Well based on some demos I have seen so far!
First of all it looks impressive, I really liked that the whole Agent workflow is in their own in-built browser which is much better than Anthropic's Computer use I feel (although that also limits the kind of tasks Operator can do as compared to the Computer use, like only being able to do task that can be done on web).
You can see the demo release here:
๐ค But what really got me thinking from these demos and from discussions on Reddit:
1. Navigating Authentication: People pointed out that there were instances when they had to manually take control and login to websites for the operator to proceed. Will we eventually need different ways to authenticate (similar to OAuth) for these kinds of agents.
2. Do we need Captcha anymore: Captcha's were introduced as a measure to identify bots/ computers and humans. But at this point I think we already know that everyone is just scraping and summarising everything on internet, and that's all done by scrapers and bots. If we know these will be doing everything on the web, do websites really need Captcha's anymore?
In both of the above cases, if I have to manually take control then that does not make the system fully autonomous.
3. ๐ฅ Decision making of agents: Saw a demo where the Operator added some products to the users Insta Mart cart, now sure it took the option it thought was best for user, but, was it?
There can be instances when users preferences wary based on their personality, may be some user like to buy cheapest option, others would want to buy the most feature rich or fresh or some brand!
๐ฅ What will be the decision making approach these agents will take? Also this bring a very crucial thought, what happens when an agent is biased to only go one route.
๐ฑ What will be the decision making approach these agents will take? Also this bring a very crucial thought, what happens when an agent is biased to only go one route.
๐ซข We live in a society where lot of our decisions are influenced based on the social media and deep down we don't like it but that's how things are.
Are we going to be a society where most of our decisions are influenced by agents and models, and who will decide and control these decisions?
I would definitely like to hear more thoughts on this.
If you have already tried the Operator, do share your perspective; I would like to hear your opinion!โ