Complete Guide to Integrating ChatGPT with Zapier: A Step-by-Step Tutorial for All Levels
Let's build with Zapier and AI (100+ videos)📅 2023-08-10
Transcript
[00:00] welcome back to web Cafe AI we do daily videos on artificial intelligence for your personal and business life in this video we're going to show you all the fundamental stuff you need to know about using the chat gbt block when it comes to automation flows so I'm going to go ahead and add this specific video to our playlist here where we dive into all 5000 apps found on zapier's backend and seeing how AI can be integrated every single one purpose of this video really is to showcase the capabilities and really understand fundamentally what hrgbt block is when it comes to Automation and all the different
[00:30] parameters associated with it one other thing I want to point out is check out the Hat here let me know in the comments if you like it if you're not familiar with that logo that is our Marketplace here at webcaf AI where we sell pre-built automation Solutions but without further Ado you'll probably see more of this but without further Ado let's go ahead and jump in and figure out everything we need to know about the chat gbt block when it comes to making zapier flows and honestly this could probably be applied to other automation platforms such as make go ahead and rename this zap to chat gbt block first thing you understand about the
[01:01] chat gbt block is it can't be used as a trigger which makes sense due to the fact that a lot of times when usage BT Block it's easier for extracting data from outputs or creating outputs using the prompt so knowing this I'm going to go ahead and just put any random trigger here we're just going to use a web hook here and jump into our chat jbt block so I'm going to type in chadbt and go ahead and start using it now one thing I want to point out here is that there is a difference between Channel gbt block and open AI blocks so the open AI block you want to use more stuff like
[01:32] uh Dalai and Whisper so let's say for example you wanted to send a prompt within chat DBT or sorry within open AIS block we hit continue here you'll see that we are limited in our capabilities at one of the major capabilities that we're limited to in an open AI block is the ability to create a memory key and as you'll see here that is pretty fundamental to a lot of automation flows so I'm going to go ahead and just delete this and let's go and jump into this Chad gbt block here and learn everything we need to understand when it comes to creating an event
[02:03] so we'll go ahead and go with conversation hit continue here it's going to connect our account out of the way I'm gonna go ahead and jump my face over here but as you see here we have a ton of different variable inputs that we have the capabilities to manipulate so let's go ahead and walk through every single one of these and tell you essentially why they are fundamental to a block let's go ahead and skip the most important one which would be the user message here and just jump into the rest of the variables to start off the first one in this context pass the user message is going to be username and assistant name this is how the data is formatted on a output so let's just go
[02:33] ahead and just give an example here we're going to go ahead and do user example and then we're gonna do an assistant um we'll put a random more like soccer and we go ahead and just test an action real quick on this so I went ahead and just changed the trigger to a scheduler so we have some data to play around with later in this tutorial but for now let's go ahead and just put in a user message of um tell me a short story so we can kind of see what this looks like for the username and this is the name we're going to come down here we're not going to add any other thing to this
[03:04] input here we're going to test action so as you see here we got the name user example and then we come down here assistant soccer these aren't too important when it comes to structuring outputs when using a DBT block I think this is more for your reference as a developer to understand the association with different outputs for now though let's go ahead and actually dive into some of the data that comes out of an output and understand everything right off the bat we have the model that's being used um the content is going to be that user message if we scroll down here there is even more information if you come down
[03:35] here a little bit more you're gonna have more important information that you're going to care about long term this is going to be the tokens used so essentially this is how many open AI tokens were associated with that prompt and output and then you have tokens remaining now tokens remaining isn't that fundamental in the context of a automation as you will not really find yourself ever exceeding that any single output due to limits that are within the the system itself so don't worry too much about token Germany so now that we understand username and user example the next one we got here is going to be
[04:05] assistant instructions assistant instructions is kind of proctoring and guiding chat gbt and giving a little context of what the fundamental task they are achieving so what I mean by that essentially is if you were making a article or you're making social media captions what you can do here essentially is say you are a social media manager give a little bit more context understand what the underlying AI output is associated with and kind of go from there the way I want you to look at assistant instructions is let's say the entire output is a banana split
[04:38] Sunday assistant instructions is just going to be the cherry on top it's really not fundamental but it can't hurt you what I suggest here though is if you're having issues sometimes mess with this as maybe this can help tailor a better output following this we have our model here so the model is probably the one of the more important things to really understand when it comes to a jgbt block because every single one of these models have very specific use cases that you're going to use them in so to start off here the gbt 3.5 model is going to be used in most contexts
[05:09] where you either want to grab data you want to basically format data you want to do stuff that is less comprehension and more comprehension oriented tasks we use gbt4 furthermore as you see here we have a gbt 3.5 turbo 16k this essentially means that when we want to deal with large amounts of data so let's say you wanted to take a 10 page or 20 Page Google doc with a ton of text maybe it's a transcript and you wanted to find the main points within it and you're having issues using 3.5 because there's errors it's loading too much too long or
[05:40] maybe the output isn't as good as you want it to be this is where you're going to transition to a 16k model finally as you see here some of these models have 0 6 13 0 3 1. you might be asking yourself what are those and what are their Association those are dates so essentially the last time that this model was up updated was March 1st in this context never use those models because of the fact that when the next update comes out these will disappear and that could cause issues within your flow so what I would suggest you to do is always use the standard gbt4 3.5
[06:13] turbo or 16k next we got our memory key here and memory key is pretty fundamental when it comes to AI Automation in general I rarely find myself never giving a memory key there may be some circumstances where it's not too important but for the majority of the time the memory key is one of the more important things that you're going to be able to use on his EBT block think of it this way a memory key essentially gives chat gbt context of previous outputs what that means for us is two
[06:44] main things first it allows us to set up flows where the output expected from chat gbt is going to be consistent due to the fact of this memory key and it knows how you want the outputs to look like so that means for us essentially we can set up a flow and never have to come back to it due to bad outputs because we set that memory key up second because of the fact that it has context for previous conversations we can ensure that content that is outputted can be unique every single time it isn't regurgitated information that's been
[07:14] said in the past this becomes valuable in content that may be in the context of you know social media content that is you know more personalized stuff that is being read by a human that you wouldn't want repeated three days later a week later and so on or a memory key is going to be a random string of 32 characters so this really can be anything so you can go ahead and put anything here as long as it's a Max of 32 characters this is how you'll proceed now one little trick here is let's say you're working with chair gbt you get an output that
[07:46] you don't like but you go back to the user message and you're like ah but this user message seems like it's prompted fine and it seems like this should be getting the right type of output one little trick here is add a one add a random character to clear the memory key and retry at the output chances are you could get a different output by just clearing the memory key that is more effective than what you originally had next we got Max tokens so in this context this has to do with output how many tokens you're willing to expand on a certain run you're going to really be
[08:16] only looking at bigger runs maybe 500 to 1000 but to be completely transparent here this isn't going to follow this to a fine T what do I mean by that essentially let's say you want an output that's 2 000 words and you put Max tokens 2 000 chances are in that Chad gbt Block it's not going to use all 2000 tokens it'll stop at maybe 400 450 and then you get a Max of maybe 400 words that came out 500 Words that came out 600 words came out this isn't an error on your end this isn't you're essentially there is no way to truly
[08:48] Proctor Chad DBT to say don't stop running until you expend XYZ tokens what does that mean for you that means that you need to essentially be able to carry on the conversation within each chat gbt block to ensure longer outputs so what that means overall when it comes to Max tokens is what I would suggest is if you're dealing with longer outputs set that to 500 set that to 750 but in reality it really is more arbitrary and you know it's not super strict on following the guidelines there now we get to our last two sections here which is going to be temperature and top P
[09:20] what this means in majority of context here is you're going to want to leave this just at one and one essentially their way of communicating what this means is that a lower top e or lower temperature is going to provide more consistent outputs and their way of basically gauging what a consistent output is is what they've seen in their data as deemed a good response and then obviously higher temperature and higher top p is going to be more quote unquote creative responses that maybe don't
[09:52] follow the guidelines of what's considered a good response but could have better responses due to the fact that they're more creative so in that context most of the time you want to leave this at one-on-one if you want to play around with it I would suggest go ahead and raise it maybe get more creativity in some context that could be pretty cool but for most use cases you can leave this at one one it's not too fundamental alright so then finally we get to our user message here just some quick rules of thumb when you do a user message dictation is everything you want
[10:23] to use as little amount of words as possible to achieve the task that you want to achieve and that's for two major reasons the first is because of the fact that every single one of those characters every single one of those words you put into that prompt is costing you money that's going to be token usage that's going to be open AI usage you don't want to spend more money if you don't need to therefore in a lot of ways of structuring prompts use as little of words as possible now one quick rule of thumb of what you can do is let's say you're trying to achieve a very specific output
[10:54] don't go with it in under the guise of okay I have to use as little words as possible proceed first with just getting the output you're looking for so if that takes you a paragraph if that takes you you know a paragraph and a half to get the exact output you're looking for okay great now we understand that whatever was supposed to be achieved got achieved then you come back to it and kind of kind of scour through it and start switching up words maybe deleting a sentence there because it actually wasn't too fundamental and so on and
[11:24] that's kind of what you need to understand for the user message if you're interested in more complex tutorials on the user message you can check out our training here at web Cafe AI or the other videos found within this playlist as we show you different ways that data can be manipulated using a user message and so on which is one side pointer here basically everything prerequisite to the chat gbt block can be inputted as data so for this example we can input the pretty date for the scheduler message here this becomes very very powerful due the fact that you're
[11:54] not dealing with fixed text when it comes to uh chat you T prompts okay that is a overall look at the gbt Block and its capabilities within AI automation you feel like you learned something make sure to like the video it's completely free and it helps us hear web Cafe AI if you want to learn more about Ai and automation when it comes to zapier and chadbt check out the playlist at the end of this video so we're diving into all 5000 apps found on Zeb use backend and seeing how we can leverage every single one make sure to subscribe if you want daily artificial content but without
[12:25] 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 web Cafe where we demystify AI for your personal and business life until next time foreign