Which AI API is Best For Creating Software: OpenAI's ChatGPT, Google's Gemini, or Anthropic's Claude
Let's become an AI prompt engineer📅 2024-02-17
Transcript
[00:00] there is a new industry that is emerging that is artificial intelligence integrated software this is going to be a industry for the next 5 years that's going to be very very prosperous if you get into it therefore a big question we must ask ourselves is which AI provider are we going to start leveraging in our software when asking this question it's important to understand the cost associated with each of these models which model performs best depending on our use case for a software and overall how to create a stable software platform with the emergence of these new AI providers welcome back y'all based off a
[00:32] video I did right here comparing the different models in the sense of just talking to it as a chat bot I did get requests to make this API version of this video so that's what I'm doing here today we're going to look at three major AI providers and I'm not even sure industry-wise if they're called AI providers I just like calling them that basically this is just when we access artificial intelligence in the context of API and integrating into our software this is a new wavelength here in the sense that you know in the last 10 to 15 years probably longer honestly ly a big way of these companies to make a lot of
[01:02] money is through backend and Cloud uh Cloud providing Google Cloud AWS making a ton of money there right but this is the next W wavelength here this is the next stage there is going to be a bunch of these big companies trying to be the AI providers for the new software companies that are coming up and will be the next million if not billion dollar company ideas we need to know which one we should go with this is probably as important as understanding which backend you'll go with with your software but that'll probably be a completely different video If I ever do create it
[01:33] but in this video we're going to dive into which API we should start accessing first major question we got to ask is the cost as we know with thousands of users the difference between a couple extra pennies per run actually is pretty fundamental when it comes to our Revenue profit and everything of this nature therefore let's go ahead and jump over to open AI right now gbt 4 is probably the most advanced model in the industry couple months from now that might be different couple years from now that might be different but as of now of this date this is the most advanced model
[02:03] that's going to cost us around one penny per th000 tokens and three pennies per th tokens output what this means in layman terms is think of every th tokens as 750 words scrolling down here though becomes you gbt 3.5 turbo which is probably going to be the more important model to talk about in today's video as when comparing the different models when it comes to just how the output looks like and the quality 3.5 turbo is going to be the best comparison when looking at Gemini and looking at anthropic so
[02:33] coming back over here understand that 3.5 turbo is 005 per th000 tokens and 015 per th000 tokens let's go and check out the other models this is Google's model called Gemini we got Gemini Pro we got Gemini Ultra as of now right now we can only access API wise Gemini Pro Gemini Pro think of being as similar to the outputs and the quality of outputs we see from 3.5 as of now we're looking at 60 queries per minute which actually isn't that high and we're looking at a price of free and a price output of free just
[03:06] because of the fact that they're training on your data but when Google AI Studio comes out these are the prices we're probably going to be caring about more with the price of 0.025 and then for an image generation 025 so you translate that that's 0.25 to 005 which is just 100% increase so the cost associated with 3.5 comparative to Gemini AP is 100% more we're talking about fractions of pennies here y'all so 100% more is not a drastic impact when we're
[03:39] looking at scale in which model we should go with therefore because we can't even access the production level API yet which is probably going to be the pay as you go when you're developing your software at the current moment in the status quo you're going to be leaning towards open AI that's going to be one of the major things going for it as they're further developed right their model is further along the track towards better and higher quality outputs because right now it seems like Google is in the stage of still they're basically still learning right that's why it's free to use but it's because
[04:10] they're training on your data they're still learning the best way that the model can communicate in the context of API coming over to anthropic we get these prices so anthropic cloth 2.1 3.5 for gbt same type of style input and output I might get some hate in the comments for that then we the pricing of 200,000 tokens for8 $8 for a million tokens what are we saying here anthropic what are we saying this doesn't make any sense like how much is it for a th000 token and I did the math for y'all the
[04:40] outcome is 0.8 or in other words basically a penny so every thousand tokens you're paying a penny the output is similar to 3.5 and we're paying only 05 here so in a way they're kind of pricing it as if the model was as good as gbt 4 Turbo from my experience the model's not there yet long term it may get there keep that in mind coming back to our whiteboard here what does that mean for us that means this first things first Gemini is
[05:14] kind of in training right now so we're just put a train this is training stage right now I feel long term this is going to be a very powerful API to access for artificial intelligence but as of now February 14th Valentine's Day or I guess when you see this will be the weekend it just seems like it's in training mode so play around with it have some fun with it but if you actually want to put a production level Branch softer ready to go paying users it's just not there yet it needs more work and thropic this has
[05:48] potential and I'm going to describe a little instance of why it has potential in the context of production level software as of now though if it's not already obvious we're going to stick with open the eye why because our loyalty lies with the most powerful model up to this point which is gbt 4 Turbo and on top of that look at the diversity when it comes to different types of models we can access for API this is an absolute fun house when it comes to creating software and more specifically the one that really gets me excited and I probably will do a
[06:20] software in the future that has some type of AI in this manner is the Vision Pro or Vision preview this is super cool y'all a lot of value there a lot software can be made using this endpoint knowing this if we come back to TL draw if we're going to go with open AI API just because of the fact that they're furthest along long term there may be a better comparative understanding of which one to choose overall I see the advancements of every single one of these as a very positive thing for the market because as you'll notice all the prices are getting slashed I gen like
[06:51] not genuinely but like I think the beginning of this entire AI thing like a year and a half ago I was worried that as we were going to continually go down this road pricing for accessing API was going to go get more expensive when it first started here it was three cents per token but now it's 1 cent per token or 1,000 tokens I was worried it was going to be like six you know I thought it was going to get more expensive but lucky for us lucky for capitalism competition has lowered the prices so you don't have to worry about cost so much as you do have to worry about the value of the output of what you're
[07:23] providing your end user therefore go of open AI now let me go aad and explain what I was referring to a little earlier when I said anthrop why would use anthropic in a production level Branch let's say we have a some logic here we got a pipeline and in this pipeline we use two prompts one two in this prompts we are using open AI what is say open AI is at o here's the thing because open AI is kind of standing out as the the one that a lot of production level software is using it's getting a lot of users or
[07:55] developers using it it does have a lot of incidents in the sense that the API does go down or just becomes worse when the API gets worse and you try to call upon it and its end points you could lead to glitches not glitches but errors within your software because it can't do that generation or can't do the output because the endpoint is just not working so how do we circumnavigate this at scale we are going to use anthropic as an insurance so you have two choices here or three choices or combination of
[08:27] choices the way we're going to do it is we're going to set up the logic that open AI kind of provides already which is called an exponential cool-off where it will retry to test or it will retry the endpoint from gbt's API and then just exponentially cool off every single time it gets like a you know bad request all right wa chill one minute okay no two minute okay no four minute and kind of proceed in that manner alternatively though if you want to still convert on that user and you're able to prompt both these prompts perfectly or maybe at some
[08:57] point in the exponential cool off we like hold up we just need to switch the model entirely maybe if it gets to like a cool off of an hour we switch to a separate model this is where anthropic could come into play this could be an insurance play in the sense that let's say there is a day where open AI API is just absolutely killed for the endpoint and if you're familiar with developing you know those days they've existed especially like for example Dev Day last year huge surge of new users couldn't even use Chad gbt for a day you know this happens it's not it's not as
[09:28] occurring as it used to be but it does happen so we need to set Insurance use it so that exponential cool off we hit an hour okay let's set up a a different type of logic here where once we hit an hour and we realize that basically this is probably going to be an issue for a couple of hours we switch to anthropic as our output AI prompt logic we have to ensure that the anthropic output is as quality as the open AI output there you go if you're going to develop a software as of February of 2024 I would suggest
[09:59] you learn open AI API documentation and proceed in that manner longterm this answer may change depending on how well Gemini performs once it's kind of out of its training stage depending on how well anthropic performs long term and everything of this nature there might even be more AI end points long term that we're not even aware of yet so saying all that proceed with open AI let me know what AI providers you're using in your software check out that playlist right there it's from concept to software I go over a bunch of details when it comes to creating software
[10:29] that's a random video that's my face I'll see you in the next video