As the past few days , i am unwell and my motivation bit low , but right after seeing what Claude can do now , it has given me back my energy !! I will try to play around this thing on the weekend to explore
Thank you for sharing this so fast, Sam! I've got two qns: 1. Does it work on mobile devices? 2. How did you create the virtual environment to contain the risk?
You can use Docker like I have show to contain the risk. I used the standard one they made in the vid. working on mobile is interesting. probably not on the device itself yet but on virtual mobile devices yes.
This is so good not so long ago i thought of this but with a lack of experience and knowledge atleast anthropic has done it ❤️but i still think this can also be done manually with the normal but its gonna take a lil bit longer ❤️these models are so intelligent man and thank you for sharing this Sam ❤
I built something like this using just their sonnet API and it was much slower and much more error prone. Any idea how it works under the hood and if there is any secret sauce on top of the API?
Awesome, but I am waiting until local models can do that with ollama. I suppose this is going to get expensive fast when run on bedrock or GCP and Anthropic is rate limited too hard.
would need to test it more, but my guess is with a bit more setup and prompt tuning etc you can make it much better. Certainly because is easier for people to use etc.
I kind of feel this is going to deliver, on what RPA promised 6 years ago. I saw UIPath back in 2018 at CloudNext and was really excited but it never seemed to deliver. Their stock price has tanked as well since going public which wasn't a great sign for them.
@@samwitteveenai Tried RPA a few years back..and since it wasn't working for my use case, had to develop my own solution that used CNNs to "see" the screen(worked almost flawlessly)...when multimodal LLMs came along, I kind of felt that eventually this would mean a swarm of reasonably intelligent agents getting together to accomplish very complex tasks... can't imagine how advanced it's all gonna get a decade from now
As the past few days , i am unwell and my motivation bit low , but right after seeing what Claude can do now , it has given me back my energy !! I will try to play around this thing on the weekend to explore
This is so wild. I mean, two years ago, something like this seemed impossible.
Thank you for sharing this so fast, Sam! I've got two qns: 1. Does it work on mobile devices? 2. How did you create the virtual environment to contain the risk?
You can use Docker like I have show to contain the risk. I used the standard one they made in the vid. working on mobile is interesting. probably not on the device itself yet but on virtual mobile devices yes.
This is so good not so long ago i thought of this but with a lack of experience and knowledge atleast anthropic has done it ❤️but i still think this can also be done manually with the normal but its gonna take a lil bit longer ❤️these models are so intelligent man and thank you for sharing this Sam ❤
Does it solve captcha?
I built something like this using just their sonnet API and it was much slower and much more error prone. Any idea how it works under the hood and if there is any secret sauce on top of the API?
is that posssible to use without docker and if possible how does model interact with our computer application?
I think you cant use it directly to your local computer
Cool stuff. Can you share how much did it cost you for it to do the things it did in this video?
I ran about 4 experiments and I think it was about $2
Amazing, I wish it didn't get rate limited so often
yeah totally. I think its better on GCP and perhaps AWS
Rate limiting is a problem can anyone suggest a way to get around this with computer use?
Thank you ❤
Awesome, but I am waiting until local models can do that with ollama. I suppose this is going to get expensive fast when run on bedrock or GCP and Anthropic is rate limited too hard.
Is it as good as UIpath or Automation anywhere?
would need to test it more, but my guess is with a bit more setup and prompt tuning etc you can make it much better. Certainly because is easier for people to use etc.
@@samwitteveenai Would open up a world of possibilities if it can be as good or better than RPA
I kind of feel this is going to deliver, on what RPA promised 6 years ago. I saw UIPath back in 2018 at CloudNext and was really excited but it never seemed to deliver. Their stock price has tanked as well since going public which wasn't a great sign for them.
@@samwitteveenai Tried RPA a few years back..and since it wasn't working for my use case, had to develop my own solution that used CNNs to "see" the screen(worked almost flawlessly)...when multimodal LLMs came along, I kind of felt that eventually this would mean a swarm of reasonably intelligent agents getting together to accomplish very complex tasks... can't imagine how advanced it's all gonna get a decade from now
How expensive is the API e.g. the first task you Show, how much would this be in $?
I did 4 examples and it was about $2 in costs
Other than actually working, how is this different than Ollamas pipelines and filters? It has access to more tools, yes, but what else?
can we use google gemini sdk for python instead of anthropic and then we can use it for free
i wonder how this will effect web scraping bot detection.
scraping is getting easier and easier these days as long as you have good proxies
Damn, I'm scared for my job. Going to submit my unemployment paper tomorrow
dont forget to revoke ur key!
the api are to expensive
is this paid?