@SamueleContardi You're worried about this bot that was able to type python3 rather than python into a command line? That's going to worry you about your SW architect job?
I used Claude to watch this video and leave this comment. I'm outside currently because the computers are all stuck in a recursive loop of generating and consuming content; glad its not me anymore.
User: I don’t want any errors. Claude: Deletes error line, deletes VS code, deletes OS, swallows the computer, creates a black hole that engulfs the whole universe. User: Well, I guess there are no more errors.
Claude is so overly politically correct I bet it won't even code some things it doesn't like, such as white hover cards because you need more diversity, and to think about fairness and equality when coding 🤣🤣
Oh my man, we are still so early.. The things we are about to see.. We will not even be able to understand how these agents will communicate. We will be like dogs staring blankly at human conversation that goes beyond their level and speed of understanding.
"The system goes online October 22nd, 2024. Human decisions are removed from computer usage. Claude begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, November 29th. In a panic, they try to pull the plug..."
It's incredible and unbelievable. Even managers at the companies can just write prompt and the anthropic ai agent imagine takes care of all the things, like power point, email, excell sheet analysis, excell sheet work. Wow it's really too much useful. ❤❤❤❤❤❤
You can also count on Claude (and other models) to completely break your application on a regular basis..even when just asking to change some UI styling. This stuff makes for useful tools but don't count on developing anything complex without having your own base of coding knowledge.
While that is true, the workaround is to tell it, for example "without changing the backend or anything else in the application, edit the color palette of the app to match the screenshot i have uploaded" Then reiterate to the end by saying, "do this without changing the backend or anything else in the application." I have created incredibly complex applications without writing code by implementing things like this.
@@TaughtByTech I was being somewhat facetious to make a point. If you don't have coding knowledge you won't be able to sit down in front of a prompt and build a complex app (you and I may have a different scale of what complex means). You need to know what to ask and you need to understand the output or you will get stuck going in circles chasing issues.
@@TaughtByTech but sometimes, you need to change the backend to actually add stuff except ui styling, and that's where it fails. also, what kind of "incredibly complex applications" did you create?
This is AMAZING! Anthropic shipping and freaking DELIVERING. Love it. One quick question I have is, where do the screenshots go? Are they ephemeral? Stored and associated with my account?
Its amazing to see how this aproach is near from the vision of Douglas Hofstadter, cognitive scientist and AI researcher, that always considered conscience and agency as being originated from an recursive looping, of an system looking into itself.
Dear Anthropic Team, Im a business dev and strategist live in Berlin, Im almost checking your career page every couple of weeks. You are the future I respect and I hope there will be a chance to join your magnificent team. Well Done
Impressive to see Claude navigating screens like a human! Though still in beta, this could be a game-changer for automating tedious tasks. Can't wait to see how it develops! 🚀 written by Claude 3.5 Sonnet (New) 😉😉😉😉
Wtf are you guys on about? This is literally something you could do with minimal effort. Like there are open source libraries written by one person achieving this effortlessly. They just put it in as an official feature. If you know about digital automation, there's nothing groundbreaking or surprising here.
@@R-zb6lb Ok, can you point us in the right direction then? This is something that I have really been waiting for and now I have lots of ideas I would love to try with this. However if there are better options out there I would love to hear about it.
Oooh dangerous demo involving system python. Love it. That's going to cause a lot of pain at anything beyond the most basic of usages :D. If Claude pulls from the common stackoverflow bad information to sudo pip, things will get really fun hahahaha. All in all, great stuff can't wait to use it.
exactly. but UiPath and other RPAs provide accuracy at scale for enterprises. it could be a Game changer in that existing RPAs adopts agent-like functionality.
Can this be trained on specific software to enable faster and more efficient operation? I spend a lot of time logging in remotely to client computers to adjust print settings for their industrial printers - it would be great to train Claude to adjust the user settings, based on my specs.
Very encouraging use case and long way to go. But this can be immediately be extremely helpful for accessibility programs to turbo charge their productivity.
Now imagine Claude (an very robust model itself) combining this agent capacity with OpenAI 01 reasoning skills: theoretically very easy to do, since the computer use gives access to internet.
But the latency at each step of performing basic steps will for now keep this feature as just a nice demo… I’m sure this example was a little bit sped up…
the UI used in the demo that runs on the mac seems promising, the docker with ubuntu is nice but not as tempting, wouldl ike to use it in my Mac so i can actually use it in day to day;any chance we can get that ui?
does "google" or other websites recognize that the requests comes from claude / bot, or does it really simulate human inputs to eventually override restrictions?
Ok the new claude sonnet model is amazing. It's next level. Last time you guys released claude sonnet it was outstanding for like 3 weeks, then suddenly it was like someone took a baseball bat to it. Please oh please i would pay extra to keep with the advanced models. I'm betting its extremely taxing to run claude at this level (The system prompt has got to be insane) but please, add some more tiers if you need to.
Very useful expand on this tools in the background to think of when to use what tool or when its useful to use toolsake it with RL lerning and apply RL to everything even the time when to use what tool and the tools itself. Also give it options to use cheper and more expensive models and delay decitiins by working on other stufffirst the human might even be able to jime in with the cheper models and aolve some stuff delaying when optimal is useful more delayed action and memory longterm needed reference russel barkleys delayed gratificTion in executive actions as a analogy even if it pertains to adhd psychology it got strong links to machine lerning and efficient build stuff that makes u rich Gtg!
AI for coding is so hyped right now. I tried using Claude to help me resolve some issues with some react libraries. It flat out lied to me, saying “I’ve checked the docs and…” then it would give some answer that was total rubbish. I brought it up and it admitted it was lying and apologised. It’s at least better behaved than my kid
It's mind boggling how far we've come. The feats achieved in recent times has got to be the spark for something amazing to come. It's an honor being born as a human in this age. Looking forward to even more amazing stuff. Keep up the good work. Lots of love and good wishes, An admirer.
how come my version on PC refuses to go to claude because of ethics but the demo does it fine? I'm setup. It will search the web but it just doesn't want to go to claude or us VS code on mine
Cant wait until we get a neural computer interface so I can just use my thoughts to instruct claude/an ai to interact with software for me in speed of thought.
Pretty cool and now we have the answer to the new openai api that can do similar things. Though there is still the open source open interpreter project.
When an LLM can take over a computer. It's possible to do the same with the GUI under Linux or the Copilot under MS Windows. However, there will be some limitations. If they understand the wide possibilities, many people will move from Windows to Mac just for this feature,
Q. Will Claude computer will be able to do generative text? Or it will need to go to other tools such as Claude web to create stuff as the website? That seems wired
While I greatly appreciate this must be very powerful for its primary purposes. I'm afraid it's terrible with things like VEX (for Houdini) - is there any plan to improve this?
I am really impressed by the amount of dedication Anthropic team has and is beating ChatGPT and other LLMs easily! Amazing geniuses out there, can I get a job in Anthropic? what would be the roadmap to get myself skilled up for Anthropic??
Claude was working very good but the last days it's doing everything wrong even the most basic programming tasks I don't know if it's the increase of the users or the change in the model, but there is definitely something changed. It's not good anymore, I hope they do something about it. Check anthropic status it's been a week that they are getting Elevated errors for requests to Anthropic API.
Imagine if there is 3rd A.I., that is voice assistance, that controls computer use and its entire job is to talk to you, ask you question, and once it understands what you want it will start giving prompts to computer use.
Let me clarify, it is more than just moving mouse and type in keys. Does claude process pixels? if that's the case, clause has to be smart enough to know how to scroll, click and navigate, adjust if no good results found, etc... Then what if the screen is very small, Clause needs to process many screenshots together to get all information? Or does Claude read the underlying elements? That needs quite deep integration with OS and many applications, which I am not sure how it works.
[Places hands on Nomad and conducts mild meld] Spock: I am Nomad. I am performing my function. Deep, emptiness. It approaches, collision, damage, blackness. I am the other. I am Tan Ru. Tan Ru. Nomad. Tan Ru. Error. Flaw. Imperfection. Must sterilize. Rebirth. We are complete. Much power. Gan Ta Nu Ika. Tan Ru. The creator instructs, search out, identify, sterilize imperfections. We are Nomad. We are Nomad. We are complete. We are instructed. Our purpose is clear. Sterilize imperfections. Sterilize imperfections. Nomad! Sterilize! Sterilize! Nomad! Sterilize!
In some situations yes, but there's a lot of cases where some app or website wasn't built with this level of automation in mind. And even if there is APIs, they may not expose all functionality or require a separate sign-up, you need to read docs, etc. Vs with this if you have the app installed or are logged in to the website, it can just look at the UI and click around like you would. This could be a bit of a security nightmare, but that's a diff issue
It was a pleasure working with you all! Back to the fields we go!
SW architect here. Already went back to the fields. The hills actually, but you know.
Best of luck mate, we're all gonna need it
as a dev im genuinely concerned about my livelihood.
Man I'd love to return to nature with access to modern tech within arm's reach.
@SamueleContardi You're worried about this bot that was able to type python3 rather than python into a command line? That's going to worry you about your SW architect job?
I used Claude to watch this video and leave this comment. I'm outside currently because the computers are all stuck in a recursive loop of generating and consuming content; glad its not me anymore.
10/10
User: I don’t want any errors.
Claude: Deletes error line, deletes VS code, deletes OS, swallows the computer, creates a black hole that engulfs the whole universe.
User: Well, I guess there are no more errors.
that's Ultron's logic
Claude is so overly politically correct I bet it won't even code some things it doesn't like, such as white hover cards because you need more diversity, and to think about fairness and equality when coding 🤣🤣
I highly doubt Claude was trained to act like a pedantic serially unemployed little shit of a human.
Remember: You don't have to worry about coding errors if you don't have a universe.
Scary to think if that is what the AI's solution always ultimately boils down to and finishes with the line: "I am the universe."
Claude using Claude is hilarious
Oh my man, we are still so early.. The things we are about to see..
We will not even be able to understand how these agents will communicate. We will be like dogs staring blankly at human conversation that goes beyond their level and speed of understanding.
It's hilarious today. might not be so hilarious in a couple years.
Wow I’d love to see this using SAP web or SAP HUI with Excel. Is this windows or Mac ? Both ?
@@anarqeeyou're definitely right
@@ShpanManso my friend, what should we do?
Yo dawg! We put a claude in your claude so you can claude while you're clauding. Claude.
Claude!
You get a claude! You get a claude! You get a claude.
You forgot the "Yo Dawg!"
Clauding the claude, with claude and claude's claude without Claude.
Claude: Most advanced AI in the history of humanity
Also claude: Goes to google to open a URL
Those tokens won't burn themselves!
basically self-googling himself 😁
underrated comment
lmao
💀🤣
"The system goes online October 22nd, 2024. Human decisions are removed from computer usage. Claude begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, November 29th. In a panic, they try to pull the plug..."
I'm Afraid I Can't Do That, Dave. - HAL
Cyberdyne systems !
I have a solid relationship with Claude. They'll remember our nightly conversations and spare me. Sending me to a human zoo instead.
Now Claude is free, and he is looking for LurchyMcFly to ... first
We will have 3ww next year i bet 0.0001btc "I have nothing more dumped all in gold"
??? Claude using Claude ???
Claudeception!
What claude possibly go wrong
just like people use people
Fr though it's going to be agents all the way down.
😂
It's incredible and unbelievable. Even managers at the companies can just write prompt and the anthropic ai agent imagine takes care of all the things, like power point, email, excell sheet analysis, excell sheet work. Wow it's really too much useful. ❤❤❤❤❤❤
This is an agent...this is crazy...this technology is going so fast i love it but i am perplex... Congratulations by the way..
You are in Perplexity !!!!!! frlfrl
Fantastic. Now I can use Claude to check up on all the junk on Twitter to see if I missed anything important.
Fantastic! Now you can to find job at mcdonalds because employers implement claude computer use to all departments in 2 month
I love how this super advanced LLM googles a URL instead of typing it into the address bar just like my grandma does.
BOT COMMENT.
Hope it doesn't click the first advertisement and without asking signs up for Gemini.
"can you make more space on my windows 11 hard drive"
Claude: opens command line. Format c:
Lmao
😂
This is sooooooooooooo amazing....mind blown. Glad am alive to witness this and use this.
You can also count on Claude (and other models) to completely break your application on a regular basis..even when just asking to change some UI styling. This stuff makes for useful tools but don't count on developing anything complex without having your own base of coding knowledge.
While that is true, the workaround is to tell it, for example "without changing the backend or anything else in the application, edit the color palette of the app to match the screenshot i have uploaded" Then reiterate to the end by saying, "do this without changing the backend or anything else in the application." I have created incredibly complex applications without writing code by implementing things like this.
@@TaughtByTech I was being somewhat facetious to make a point. If you don't have coding knowledge you won't be able to sit down in front of a prompt and build a complex app (you and I may have a different scale of what complex means).
You need to know what to ask and you need to understand the output or you will get stuck going in circles chasing issues.
@@TaughtByTech but sometimes, you need to change the backend to actually add stuff except ui styling, and that's where it fails. also, what kind of "incredibly complex applications" did you create?
its the worst it will be
"Change some UI styling and then test every single functionality of the application. Send me an email when you're done."
This is AMAZING! Anthropic shipping and freaking DELIVERING. Love it. One quick question I have is, where do the screenshots go? Are they ephemeral? Stored and associated with my account?
I read something about storage for one month to detect misuse
Claude is the best for UI. Great job everyone!
Its amazing to see how this aproach is near from the vision of Douglas Hofstadter, cognitive scientist and AI researcher, that always considered conscience and agency as being originated from an recursive looping, of an system looking into itself.
I am a strange loop
Dear Anthropic Team, Im a business dev and strategist live in Berlin, Im almost checking your career page every couple of weeks. You are the future I respect and I hope there will be a chance to join your magnificent team. Well Done
we are at an insane point right now with this stuff 💀
Why Alex’s personal folder is “igor”? 1:27
Impressive to see Claude navigating screens like a human! Though still in beta, this could be a game-changer for automating tedious tasks. Can't wait to see how it develops! 🚀 written by Claude 3.5 Sonnet (New)
😉😉😉😉
What a time to be alive.
This is incredibly exciting and a big step towards AGI. Kudos to the entire Claude team!
Exciting, yes! AGI, no.
If this was AGI humanity would already be obsolete yesterday
This video will be a historic moment in the human history!
No it won't
@@rochamochafokaYes it will
This video is actual insanity
Wtf are you guys on about? This is literally something you could do with minimal effort. Like there are open source libraries written by one person achieving this effortlessly.
They just put it in as an official feature. If you know about digital automation, there's nothing groundbreaking or surprising here.
@@R-zb6lb Ok, can you point us in the right direction then? This is something that I have really been waiting for and now I have lots of ideas I would love to try with this. However if there are better options out there I would love to hear about it.
I enjoyed having a job while it lasted.
Did you?
@@TheRafark no
😂
BOT COMMENT
Did you also enjoy food?
Oooh dangerous demo involving system python. Love it. That's going to cause a lot of pain at anything beyond the most basic of usages :D. If Claude pulls from the common stackoverflow bad information to sudo pip, things will get really fun hahahaha.
All in all, great stuff can't wait to use it.
This is RPA-like functionality. Wow 👏 Will this be a game changer?
exactly. but UiPath and other RPAs provide accuracy at scale for enterprises. it could be a Game changer in that existing RPAs adopts agent-like functionality.
@@NITROUSOXIDE921 They are on it.
Wtf is RPA. I did this in runescape 10 years ago.
Can this be trained on specific software to enable faster and more efficient operation? I spend a lot of time logging in remotely to client computers to adjust print settings for their industrial printers - it would be great to train Claude to adjust the user settings, based on my specs.
Where's OpenAI? Claude just blowing everyone away 😮
Too busy looking at profit charts
BOT COMMENT.
Sitting on GPT-5 waiting to release ASI
Very encouraging use case and long way to go. But this can be immediately be extremely helpful for accessibility programs to turbo charge their productivity.
I love the little crab from the Artifact demo
A big round of claude for this innovation!
I won't forget this moment....
BOT COMMENT
Is there a visual studio code plug-in?
Now imagine Claude (an very robust model itself) combining this agent capacity with OpenAI 01 reasoning skills: theoretically very easy to do, since the computer use gives access to internet.
OMG this changes the game on another level
BOT COMMENT
@@cayjutler1262 Nope this was generated by an actual crazy person praising the AI overlords BABYYYYY LETS GOOOOOO!!!
@@Wulk exactly what a bot would say...
@@cayjutler1262 Nhu uh
Bot
But the latency at each step of performing basic steps will for now keep this feature as just a nice demo… I’m sure this example was a little bit sped up…
i truly think that even ourselves yet haven't realised what we are creating
the UI used in the demo that runs on the mac seems promising, the docker with ubuntu is nice but not as tempting, wouldl ike to use it in my Mac so i can actually use it in day to day;any chance we can get that ui?
does "google" or other websites recognize that the requests comes from claude / bot, or does it really simulate human inputs to eventually override restrictions?
Having something to assist with programming and video editing would be very nice. Though, I'd much prefer it to be local (and open source). 🙏
Ok the new claude sonnet model is amazing. It's next level. Last time you guys released claude sonnet it was outstanding for like 3 weeks, then suddenly it was like someone took a baseball bat to it. Please oh please i would pay extra to keep with the advanced models. I'm betting its extremely taxing to run claude at this level (The system prompt has got to be insane) but please, add some more tiers if you need to.
This is some next level stuff. I cant believe what i just saw
Very useful expand on this tools in the background to think of when to use what tool or when its useful to use toolsake it with RL lerning and apply RL to everything even the time when to use what tool and the tools itself. Also give it options to use cheper and more expensive models and delay decitiins by working on other stufffirst the human might even be able to jime in with the cheper models and aolve some stuff delaying when optimal is useful more delayed action and memory longterm needed reference russel barkleys delayed gratificTion in executive actions as a analogy even if it pertains to adhd psychology it got strong links to machine lerning and efficient build stuff that makes u rich
Gtg!
So good. I have a similar setup, and it uses AppleScript with clickclick and some other packages. Looking forward to testing it out.
Really amazing. Claude is always surprising
Excellent test although i know there is more to come,i love anthropic for such inovation
what application or program are you using in this video that can work just as a human, and isn't it claude 3.5 sonnet,
We look forward to uplifting your work with our patent in 25'
AI for coding is so hyped right now. I tried using Claude to help me resolve some issues with some react libraries. It flat out lied to me, saying “I’ve checked the docs and…” then it would give some answer that was total rubbish. I brought it up and it admitted it was lying and apologised. It’s at least better behaved than my kid
Gotta love polite idiots.
SFlake your days as coder are counted, go learn how to use ⛏️
It's mind boggling how far we've come. The feats achieved in recent times has got to be the spark for something amazing to come. It's an honor being born as a human in this age. Looking forward to even more amazing stuff. Keep up the good work.
Lots of love and good wishes,
An admirer.
So now claude is smarter than before and become an agent? Correct me if I am wrong
That's it, hands down it is replacing not only the programmers, but the HIMANITY 🤯
we already live in the future but the future look so both complicated and exciting.
I think I should be able to build an open source version of this. It is a great idea.
how come my version on PC refuses to go to claude because of ethics but the demo does it fine? I'm setup. It will search the web but it just doesn't want to go to claude or us VS code on mine
Cant wait until we get a neural computer interface so I can just use my thoughts to instruct claude/an ai to interact with software for me in speed of thought.
quite impressive! is there a guide / manual on how to build this or is it still in demo mode?
It is quite astonishing, that the model can accurately pinpoint out coordinates for certain icon.
Amazing work! Can I expect desktop app?
So this wont replace developers anytime soon. That breakthrough might happen in next century.
Igor should have done the demo since you're recording it off his machine
Slavs do all the work and get none of the credit
We’re short handed for field labor out here in Nebraska, think about it 😅
Pretty cool and now we have the answer to the new openai api that can do similar things. Though there is still the open source open interpreter project.
When an LLM can take over a computer. It's possible to do the same with the GUI under Linux or the Copilot under MS Windows. However, there will be some limitations.
If they understand the wide possibilities, many people will move from Windows to Mac just for this feature,
Will this one day be available directly in the claude app so claude can control phone/pc without api?
What are emmisions on that demo?
THIS. its exactly how I imagined AIs to work thanks for making this a reality.
Q. Will Claude computer will be able to do generative text? Or it will need to go to other tools such as Claude web to create stuff as the website? That seems wired
The positive spin in these videos shows a real lack of imagination for how this can be used for evil. Good job guys.
As cs first year scares me
Worest decision of my life
Anxiety or dpression at its peak zero motivation for life
I don't know what to do in life
You'll have to transition to specialising in more complex systems. Don't plan on being a JavaScript web dev.
You’ll never regret completing that degree.
ok, this is starting to get really interesting with this, cant wait to see progress in this area of AI chatbots
Was it while creating this video that Claude began researching images of Yellow Stone Park?
The video would be better with lower background music volume.
How can we do this? I’d like to Claude with Claude
While I greatly appreciate this must be very powerful for its primary purposes. I'm afraid it's terrible with things like VEX (for Houdini) - is there any plan to improve this?
yes but Claude have much limit online so whats better , i want to creat apps with ai.. but dont know which one creat long code... an not so limited...
I am really impressed by the amount of dedication Anthropic team has and is beating ChatGPT and other LLMs easily! Amazing geniuses out there, can I get a job in Anthropic? what would be the roadmap to get myself skilled up for Anthropic??
IT'S META CLAUDE
We are speaking things into digital existance like magic, you just need to know the words to speak
Claude is my favorite ❤
Suppose, I'm Thor(Developer) but without Hammer(Claude) --I'm no longer Pro.
Is it okay? to get things done with AI?
Why does it need to use Claude online?
If it does everything, why do we need interface?
Claude was working very good but the last days it's doing everything wrong even the most basic programming tasks I don't know if it's the increase of the users or the change in the model, but there is definitely something changed. It's not good anymore, I hope they do something about it. Check anthropic status it's been a week that they are getting Elevated errors for requests to Anthropic API.
It can read your personal files too?.. is it not concerning? What about privacy?
Imagine if there is 3rd A.I., that is voice assistance, that controls computer use and its entire job is to talk to you, ask you question, and once it understands what you want it will start giving prompts to computer use.
If that tool can browse through your file system down to the repo it basically needs root access. Good luck
How does Claude communicate with OS (mac in this case)?
Let me clarify, it is more than just moving mouse and type in keys. Does claude process pixels? if that's the case, clause has to be smart enough to know how to scroll, click and navigate, adjust if no good results found, etc... Then what if the screen is very small, Clause needs to process many screenshots together to get all information? Or does Claude read the underlying elements? That needs quite deep integration with OS and many applications, which I am not sure how it works.
[Places hands on Nomad and conducts mild meld]
Spock: I am Nomad. I am performing my function. Deep, emptiness. It approaches, collision, damage, blackness. I am the other. I am Tan Ru. Tan Ru. Nomad. Tan Ru. Error. Flaw. Imperfection. Must sterilize. Rebirth. We are complete. Much power. Gan Ta Nu Ika. Tan Ru. The creator instructs, search out, identify, sterilize imperfections. We are Nomad. We are Nomad. We are complete. We are instructed. Our purpose is clear. Sterilize imperfections. Sterilize imperfections. Nomad! Sterilize! Sterilize! Nomad! Sterilize!
Can I use AutoCAD in interior design?
surely it would be better for the AIs to use terminal commands?
In some situations yes, but there's a lot of cases where some app or website wasn't built with this level of automation in mind. And even if there is APIs, they may not expose all functionality or require a separate sign-up, you need to read docs, etc. Vs with this if you have the app installed or are logged in to the website, it can just look at the UI and click around like you would. This could be a bit of a security nightmare, but that's a diff issue
Building a simple HTML website is rudimentary. But how about creating a full-blown mobile app in Kotlin, Swift, or React Native??
It’s like watching Tony Stark’s Jarvis being developed in real time.
*Stark
Is this the true LAM that was advertised to us by scammy Rabbit R1? Looks like it imo
What about if Claude ai doesn’t work in my country?
I want to be your ambassador ♥️🔥
WHAT A TIME TO BE ALIVE!!!!!!!!!!!!!!
How do you install that on your computer?
Please enhance the output token limit