I got the new autonomous agents from anthropic up and running in a demo so you didn’t have to.

Description

Chapters

Transcript

0:01 Have you heard of the new anthropic
0:02 computer use demo where it can use your
0:05 computer? It's crazy. I got it up and
0:08 running and I'll tell you about it. This
0:09 is Kyle Shannon with the AI learning
0:10 lab. Uh let me show you a little bit
0:12 about what's going on. Okay, so I have
0:15 the computer use demo up and running. So
0:18 this is the little chat window or chat
0:20 yeah chat uh history. And then over here
0:24 is a browser. So, I'm running a virtual
0:28 environment, which I'll talk about in a
0:29 second, and then I'll tell you what it
0:31 did in a second. But first, how do you
0:34 get this thing running? It's It's a pain
0:36 in the ass. And so, if you don't feel
0:39 like geeking out, just you can just
0:41 watch I'll show at the end of this what
0:43 it actually did, a demo that I did for
0:44 myself just to see if it would work. And
0:46 it did. So, that's cool. Um, what you
0:50 have to do or what I Here's what I what
0:52 I did. What I did, I found a YouTube
0:55 video. Uh there's a bunch of them out
0:57 there. Just find someone who doesn't
0:59 annoy you. Um and then what you have to
1:02 do is you have to download this app
1:04 called Docker. And what Docker does is
1:06 it creates a virtual computer inside
1:08 your computer so that you don't run this
1:12 thing and like inadvertently erase your
1:15 hard drive or something stupid like
1:16 that. So, so Docker allows you to have
1:20 like a fake computer that runs a browser
1:23 and it's got a spreadsheet program in it
1:25 and things like that. Um, you have to do
1:28 some configuration. So, you have to have
1:30 a claude or an anthropic API key. So,
1:34 you have to have a developer account.
1:35 So, if you go to console.anthropic.com,
1:39 that's the developer thing. That's where
1:41 you get an API key. Give them a credit
1:43 card so you know you can run the API.
1:47 Um, and then basically you need to get
1:50 this thing running in your terminal on a
1:52 Mac and a similar a console on a PC I
1:55 think. Um, basically you get Docker
1:58 running in there and you go to there's a
2:01 GitHub repository
2:03 for for um for the computer use demo. So
2:08 just search for GitHub anthropic
2:11 computer use demo and then there's a
2:13 little chunk of code for Docker and you
2:15 basically just paste that in your
2:17 console and hit run. Um for Docker you
2:20 do have to have your account connected
2:23 or it won't work. It'll look like it's
2:25 working but it won't work. That took me
2:26 a while to figure out. The whole thing
2:29 probably took me an hour because I'm
2:31 totally clueless. If you kind of know
2:33 what's going on with Docker and Terminal
2:35 and stuff like that, it's probably 1015
2:37 minutes. So, it's not that big a deal.
2:39 But let me show you what it actually did
2:43 prompt that I gave it and then what it
2:45 did. And it's pretty cool. Okay. So, the
2:48 first test I ran, I just had it ask uh
2:51 who the CEO of StoryVine was. And it so
2:54 what it first does is it has a uh it
2:58 takes a screenshot of the computer
3:02 window the virtual computer window. So
3:03 it initially did not have a browser
3:06 running. It then launched Firefox. It
3:08 then went out and it searched for uh
3:12 Storyine CEO and Google just gave it to
3:14 it. So it was really quick. So then what
3:16 I asked it and this was this was a bit
3:18 tougher. I said I like Martin Ston. find
3:21 me music in Denver this weekend. Um, in
3:24 a similar style and so it went to the
3:26 browser, it screwed something up and got
3:29 an error. Um, but it saw that it took
3:32 the screenshot and then it started
3:34 typing in things about Martin Ston or or
3:38 a certain kind of music and then it
3:40 found some things in Denver and then it
3:45 it said, "I found a few events this
3:47 weekend that might appeal to a Martin
3:48 Ston fan. Here's what I found. Geobbard
3:51 zero's power of music. And then it said,
3:53 "Let me check bands in town for more
3:55 options." So then it went back and just
3:57 started, you know, searching bands in
3:59 town. And then at the very end, it gave
4:02 me this really nice write up. After
4:04 searching through various venues, here's
4:07 the two things. Earlier in my search, I
4:09 noticed Ellis Paul is performing. Um,
4:13 Ellis Paula is actually a great match
4:16 for Martin Sexton fans as he's also from
4:18 Boston folk scene and shares a similar
4:21 folk rock storytelling qualities and
4:23 strong vocals. So, it understood who
4:25 Martin Sexton was and it just did all
4:27 this stuff, you know, just taking
4:29 screenshots and prompting itself to go
4:32 solve everything. So, if you want to
4:35 geek out, it's not that big a deal. Like
4:38 when I tried to set up stable diffusion
4:39 when I first got started, I failed nine
4:41 times. This thing I only failed like
4:44 three times. So So it's it's it's not
4:46 that tough to set up, but you definitely
4:48 have to geek out a little bit.
4:51 If you don't feel like geeking out,
4:53 there's probably stuff coming probably
4:55 within a month where all of this will
4:57 just be built into like chat GPT or
5:00 clawed where you don't have to do all
5:01 this programming [ __ ] to get it to work.
5:03 But if you want a taste of the future,
5:05 if you want a taste of what agentic
5:08 behavior is, autonomous agents taking
5:11 action our on our behalf, it's really
5:13 fascinating to watch this thing work. I
5:15 did this was a really simple example.
5:17 So, it didn't have a lot of, you know,
5:19 chance to fail. So, I've seen a lot of
5:21 people try to get this to work. It
5:23 absolutely fails. But this is the
5:25 future, people. Agents are the next big
5:28 thing. They're here and they're going to
5:30 get much better really soon. So hold on
5:32 to your butts.