Assistant API No-Code Integration: OpenAI Meets Zapier For Smart Automations - Complete Guide
Let's build with Zapier and AI (100+ videos)📅 2023-11-27
Transcript
[00:00] can we put assistance API into zapier Now using no code yes your first question might be what is assistance API assistance API think of it as hiring a human for customer service or a specific use case that actually has memory in an automation EG you don't have to do anything once you set it up this is extremely powerful I don't even know why zapier didn't publicize this maybe they did maybe I was under a rock but we can now create assistance in zapier with no code so we're going to jump into that today welcome back y'all to Corbin AI we we are diving into artificial
[00:31] intelligence together these videos are more of my automation business focused videos let's go jump in here this is crazy um so I was playing around with zapier yesterday I was kind of like grabbing and messing with some my own automations and I noticed and I noticed that there was a option to create an assistant so coming over to my Twitter here I went ahead and just published it I was like uh yeah so it seems like zapier now can create assistance and I wasn't aware of this I wasn't no one sent me a message I didn't see this
[01:01] anywhere in zapier's uh publicy or Publications this is huge y'all this is like they basically broke a huge barrier where you don't even need a code to access the assistant API anymore and the assistant API is extremely powerful y'all therefore in this video we're just going to I'm just going to show you what it does and how you can start leveraging it I'm going to give different context and how you can approach it so we're going to do a good oldfashioned Google Drive trigger here because what we're going to do here is we're going to analyze a PDF using an assistance API also I'm going to show you how it looks
[01:32] in the uh open AI dashboard and backend for now though we're just going to say new file in folder and continue here I'm going to choose our courses account here hit continue again and then we're going to choose our YouTube folder here that we do YouTube test apps on this channel here I'm going to go in here I'm going to say new feature and then I'm just going to proceed from there so what I did already is I have loaded a example PDF here yes it has to do with climate change and global warming again the reason I do that is cuz these the only types of PDFs you can get online without having to put in an optin it's just a
[02:03] free pdf purpose of this is purely just for data point being though is now we have this huge PDF right bunch of data bunch of you know information that AI can now you know look at and internalize knowing this we're going to go ahead and create an assistant together and understand the functionality of an assistant so I'm going to go and just rename this zap to assistant creation I'm going to need to make more videos on this topic honestly when I was just playing around with it on my own I was like oh smokes there is so much you can do now it is actually this is this is
[02:34] next level here this is kind of like you don't really need to do customer service to an extent if you really know how to build out these correctly knowing that we're going to go and test this trigger here we're going to walk through everything together here so I'm going to test this trigger here this is just our pseudo file a little global warming report or PDF research and let's go ahe and do it so we're going to go ahead and do chat gbt here and to start off we are just going to do an event of creating a assistant saying that we can if I can find it yeah so we can go into open AI dashboard here and we can create it and then reference it but rather than doing
[03:05] that I want to go and create it with y'all in zapier itself because the front end is nice and you can kind of get more perspective so we're going to go ahead and create an assistant here so we're going hit create assistant continue and we're going to just fill in some information that we care about here so the first thing is going to be the model right now assistants only work with the most up-to-date models of gbt 4 and the most updated model of gbt 3.5 what I mean by that is the the one that has 1106 by it and the gbt 4 model that has 1106 yeah right here gbt 4 1106 preview
[03:38] knowing that here is what you need to keep in mind first thing you need to keep in mind gbt okay let me just take a step back here gbt 3.5 turbo and gbt 3.56k the way they were differentiated was 16k could handle more data right 16k in that context is 16,000 tokens I believe and basically can handle more data now what they've done with the model is now GB BT 3.5 turbo the new model that they're pointing towards when they say 116 by the way that means November 6 that new model is a by default 16k when comparative older 3.5
[04:11] models was set at 4K that might have sounded like a lot of gibberish all you need to know is that 3.5 turbo 116k is cheaper and basically better so I'm going to go and click that the way we want to use 3.5 turbo in this context is we're going to use it just to analyze large amounts of data one other thing I want to point out though which is really cool y'all so when you're building these resistance apis ah I got a PDF I got it's got images or it's got you know Excel sheets whatever whatever don't have to worry as this assistant API can do gbt 4 1106 preview this probably
[04:42] would sound like gibberish six years ago what are you even saying these are just random characters is that is that a lottery number no um that can visualize PDFs but I'm jumping ahead of myself we're going to go and do gbt 3.56k we're going to go and just name this for our specific use case I'm just going to call it Global scientist and from here we got a description we can give the underlying assistant we got the instructions for the assistant we got the tools for the assistant this is where it gets cool retrieval and code interpreter so
[05:13] obviously if you're dealing more uh data Centric math Centric type stuff you do code interpreter you can enable both but for now all we care about is the retrieval tool of you know being able to ask questions about a PDF boom just blew up blew someone's brain right now because they're just like they've been asking me for this how do I do this there was ways I showed you how to do it there's complex ways to accessing this kind of data now it's just simply adding it as a tool really cool then we have new files or existing files so when you're initially creating these
[05:43] assistants we are we can frontload it with just data that we want that's going to be un unilaterally used throughout your automations so in this context since it's a global scientist we're going to load it with a the data of that PDF saying that that the way you would approach this as you'll see as we continue this tutorial together is we can provide new files for specific context on top of that these files that we load with the assistant API would be prevalent information about your business EG customer service you may
[06:16] have a PDF that says frequently asked questions or ways or templates or ways of approaching certain type of inquiries you would load that PDF from here though we're going to go ahead and continue here and we're going to go ahead and create our assistant once our assistant created we'll actually be able to see it in the dashboard of open the ey that being said if you've gotten this far and you haven't used zapier or you haven't use open a and you're just like this is cool check out that video right there it's a 40 minute plus video just shows you how to even integrate both the softwares and stuff you should just know as a general basis but from now going to
[06:47] assistance API here if I hit reload we're going to see global scientist boom this is really cool y'all no code so far from here we can go ahead and start understanding what what even is going on here so I'm going to add another Chad gbt block here and the way this works and why this is so powerful is notice how we have two different options now we have conversation and conversation with assistant conversation is what we've been using for the past eight nine months together and conversation and it's good it's solid there was ways to go about it to provide
[07:17] memory for previous conversations for conversation this is sound crazy but there was way to go ways to go about it but it required a lot of just you know troubleshooting conversations with assistant now we can really lock in so I'm going to take conversation with assistant the way we're going to approach this is this first thing I want to identify is we need to identify an assistant now typically at scale you won't be having a function of creating an assistant in a flow because that would basically mean if this flow
[07:48] triggered 20 times we had 20 different Global scientists I just did that for the purpose of just you know doting my eyes crossing my tees so in the context that we don't typically do that the way we would approach it is for assistant and rather than doing you know the easy thing well actually in theory you could do the easy thing of clicking it here EG and knows in our back end that it can grab it here also on top alternatively you could just grab the underlying ID associated with it and set it up as custom and then put the ID here for us though it's going to show up anyways and
[08:19] in the context that this is a customer service agent it would just say customer service agent so we already know that customer service agent no matter where we place it and all of our automations is front loaded with the data that we provided knowing that we have a couple of different metrics here that we can start filling out name instructions this could be pertinent we'll go ahead do that together model this may be more important to you so the way I want you to structure your assistance is I want you to use 16k I want you to use 3.5 turbo when you're front loading that much data in the
[08:51] beginning because I want you to be able to compress it so it's not too big for uh gbt 4 that being said gbt 4 can still do it it it's just put it this way if your data can be interpreted and analyzed and you have a lot of data use the 3.5 turbo it's cheaper and it can handle more but that being said gbt 4 still can handle a lot more as well the window got expanded as well so therefore if you're using IM just play around use either one I would lean towards 3.5 in the beginning for for more complex stuff
[09:22] but for now coming over back to conversation with assistant here's what we can do here so conversation ID is the money bag right here this is kind of like the better version of a memory key why is it better than a memory key because each single a memory key was more Broad in the sense of like consistent output consistent output consistent output consistent output based off previous outputs this one gets a little bit crazier because what we can do with this one is that now we're getting instead of we're getting consistent outputs we're getting outputs that are personalized to every single
[09:52] interaction individual has with the automation comparative to just having a social media caption with no uh uh quotation marks that being said let's just show you in action so I can you know so you can kind of see what's going on here so knowing that you can also provide existing files in the flow so basically on top of the PDF I Pro provided by global warming I could provide another PDF on top of that we can add a new file to the flow or sorry to the assistant response instructions and we can even add maybe want a code interpreter for a specific use case on
[10:22] top of having the ability to read PFS knowing all of that let's go and proceed we're going to say let's actually jump over to the PDF real quick so what you'll notice is is Big PDF but what you'll notice is that in the right off the Bates fossil fuels provide 85% of the world's energy so we're going to basically gut check this and make sure you can even read the PDF we're going to say what uh percent of the world's energy comes from fossil fuels so this is just our way of gut checking to see you know if it's
[10:55] going to say if it says 85% we know it to be true and it actually internalize the data we provided so I'm going hit continue here and I'm going to test this step there we go so as you see here we got a full response 85% so we know we can actually internalize and read a PDF now here is where it gets super cool y'all so let's go ahead and play around with that conversation ID this is basically when I saw this in CER I was like okay now we're at Next Level here now we can really personalize automations to the point that you may
[11:25] you may not even be able to differentiate between a human and a robot now obviously Actually I don't even know now they have the ability to do text to speech so we incorporate that as well like you could get really crazy here saying that though basically here's what we need to know what that event occurred was after I had a conversation with that chat gbt here as you'll see to the right here we have a message ID so it's kind of identified as message ID message ID goes into conversation ID why they named it differently I don't know
[11:57] but this is kind of like like the the specific conversation that occurred there EG the message that happened it knows it happened so to prove this let's go ahead and do that but here's the thing very very the very first inquiry is what creates the conversation ID therefore we won't actually know the very very first thing that the individual says but everything past that we can reference so what I mean by that is if I asked it right now what was just
[12:28] said said it wouldn't know because in this first initial step here we didn't have a conversation ID created yet but after the step incurred then we had it created therefore watch this so I'm ask the same question here continue and notice how I have a message ID now I have the memory key of that specific message here so the first thing it's clicking into is okay it just ask what percent of the world's fossil fuels coming down here 85% perfect let's gut check it what did I just ask
[12:59] notice how I'm not adding anything else I'm just asking what did I just ask so if we use conversation the traditional way of approaching Chad gbt it would have no clue but with that message ID it's going to have a clue Blues Clues y'all remember that show where like they're hiding little blue paw prints and stuff and then we got mail okay uh you asked about the percentage of the world's energy that comes from FAL fuel I found it in the document boom that
[13:31] right there just I know at least half of yall that are watching right now some Wheels were turning your head and it just clicked you're just like oh okay now I understand why Corbin was so ecstatic in the first half of this video because the first half of this video I just think he was tripping we're not tripping here this is super cool y'all we can do a ton of stuff with that like so much stuff that when I did that I was like okay this is um like I still want to diversify my content but man this D wrenching it cuz I'm like oh my gosh there is so many videos we can do on that capability
[14:02] alone and this is just scraping the surface so knowing that we now know that we can reference I so let me just throw out a crazy idea here an idea that I plan on doing as a future video so make sure to leave a like if you felt like you learned something so far so one of my past videos here I've done is ability to personalize marketing using zap your's tables what is zap your tables AB tables think of it like an Excel sheet Google Sheets where we store data on individual with a unique identifier therefore we can reference them later if you're a developer think of Firebase backend AWS backend stuff of this nature
[14:33] like fir store right so knowing this though we can now store specific conversations with said individual therefore we can personalize how we interact with that individual even further furthermore Corbin I people be going through my flows I want to know what they're saying to my robot or I want I know I want to know their conversation I want want to know what's incurring here now we could save this chat as well so there is a lot
[15:04] of stuff I mean man this has changed a lot in the past 8 to 10 months of what we can do here so we have been taking Leaps and Bounds when it comes to artificial intelligence integrated with automation so saying all that though that kind of concludes this video I know sneak peek you're like come on give me give me some more give me a cooler give me a cool future is we're getting there don't worry we're going to get there cuz this just basically opened Pandora's Box because the alternatively when this first was released a big downside was
[15:36] you needed to have some threshold of understanding how to code and on top of that who the heck wants to go out of their way to learn Google Google Drives API documentation set up on VSS code code all code this entire Pro no one wants that give me the drag Dragon Block I'm playing Legos out here all right if you feel like learned something so far or to the end of this video make sure to leave a like helps me out here if you want to see more tutorials like this when it comes to zapier and AI you can check out a playlist at the end here where we're diving into all 5,000 apps I'm just showing you ways to start
[16:06] leveraging it uh this is exciting stuff I plan on doing more stuff on this topic make sure to follow my Twitter and Tik Tok but without further Ado I'll see you in the next video thanks for tuning in and yes surprise I'm an AI Avatar make sure to explore more here at Corbin AI where we demystify AI for your personal and business life until next time