Task: Please review the following video transcript and perform a comprehensive edit to…
- User
Task: Please review the following video transcript and perform a comprehensive edit to ensure it meets professional standards.
Instructions:
- Correct Grammar, Punctuation, and Spelling:
- Identify and fix any errors in grammar, punctuation, and spelling throughout the transcript.
- Ensure that sentences are structured correctly and convey the intended meaning clearly.
- Verify Terminology and Concepts:
- Check that all technical terms and concepts relevant to the video's subject matter are accurately used.
- Replace any incorrect or ambiguous terminology with appropriate and accurate language.
- Ensure consistency in the use of terms throughout the transcript.
- Enhance Clarity and Precision:
- Rephrase any awkward or unclear sentences to improve readability.
- Eliminate unnecessary repetition or redundancy.
- Maintain the original meaning and intent behind each statement.
- Preserve Original Meaning and Tone:
- Retain the speaker's original tone and style to the extent possible.
- Ensure that the edits do not alter the intended message or nuances conveyed in the transcript.
- Format for Readability (Optional):
- Organize the content into clear paragraphs.
- Use headings or subheadings if it helps improve the structure.
- Consider adding bullet points for lists or key points to enhance readability.
Objective: The final edited transcript should be a polished, accurate representation of the original video content, easy to read and understand for the intended audience. """ Title: "How to Build AI-Powered Apps with OpenAI \u0026 ChatGPT (That Don’t Sck)" Transcript: "(00:00) The integration itself isn’t that complicated. The way I see it, if you don’t have technical knowledge… You can do it with Zapier, or You can do it with a bit of coding. I’ll show you. Trickier, but some advantages. What I really want to talk about is if anybody can do it, should you? I promise, no nagging. (00:20) Oh wait I apologize, there will be nagging. First, let’s give you what you came here for. How to build apps that revolve around OpenAI, or how to integrate OpenAI into an existing app. Second, we’ll look at how OpenAI has made AI app development really easy, and why that’s not necessarily a good thing for would-be founders. (00:41) Just to be clear, when people say integrate ‘ChatGPT’, sometimes, what they really mean is ‘integrate an ‘OpenAI model’. There are a few, and ChatGPT is an example of an app built with an Open AI model. This one just happens to specialize in natural language production and surface level domain expertise. (01:05) Emphasis on surface level, more on that later. If you want to create an app powered by an OpenAI model, you need to know your use case available OpenAI models their specializations, and cost - they’re not prohibitively expensive, but they’re not free Ahh… You’ve been using chatGPT and thought it was all free? Nope! It’s a free… sample. (01:26) You want the real deal, it’s going to cost you. Here’s a list of available models together with cost. GPT-4 - GPT 3.5 on steroids GPT-3.5 - understand and generate natural language/code DALL·E - generate and edit images Whisper - convert audio to text Embeddings - Convert text to numerical form I might have missed a few - there are a lot of them - but basically, you’ve got: Language models Image models, and Audio models All of them have their place. (01:55) I know I’ve said it, but it’s important and bears repeating. It comes down to your requirements. If you don’t know, two options: One, look at the models and try to find use cases that fit Two, look for a use case, then find a model that fits These are the steps to integrate an Open AI model into your app: Get a unique OpenAI API key by signing up with OpenAI. (02:21) Construct a prompt and a frontend for users to interact with. Include specific instructions and necessary context. Normally, this means static instructions accompanied with dynamic user input. Sounds unnecessarily complicated. Basically, you create a template, and your user fills in the blanks. Super simple example: Two text boxes on a screen. (02:43) One text box contains the prompt or static instructions and user input which changes with every request. Here’s an example of an input and output area in a Google Sheets conencted to OpenAI via Zapier. Here’s an example of a very basic app interface built from Locofy’s training template. The input in the first box is part of the prompt sent to OpenAI. (03:07) OpenAI creates a completion. The second text box receives that completion. Evaluate the response: Was it what your users would have wanted? Adjust the prompt and test until you get accurate and reliable results. Step 1 and 3 are pretty straightforward. Step 2 is the tricky one. And there are 2 ways to do this if you’re someone without coding knowledge. (03:27) The first, and easier way is through Zapier. Zapier is a SaaS that simplifies data transfer between other apps. It’s got common data transfer APIs built-in - you just arrange what you want. It reminds me of this game called Rocket Mania. You’ve got a match on one end and a rocket on the other, and you arrange fuse blocks to establish the connection. (03:50) Same with Zapier. You don’t have to build anything, just arrange what’s already there and let the data flow. Good news: Zapier has prebuilt connections with thousands of apps. I got my writer to try this method out, and he has zero coding knowledge. Like, if I had to rate his coding out of ten, it would be scks balls out of ten. (04:08) He wrote that by the way. I saw no reason to disagree. He used Google Sheets as his frontend…and lets be charitable and give him that. If you put aside how it looks, it’s an app. And his app is for people to get information on African cichlids. What the hell is that? African cichlids are a group of freshwater fish from the African continent. They’re really colourful fish and popular aquarium pets. (04:31) I’m not a fish guy, so I asked Ashraf to explain his use case, and this is what he wrote: In fishkeeping, fishes can have many common names which leads to confusion among beginners and even intermediates. However, there can only be one current scientific name, so fishkeepers prefer using that when doing research to reduce risk of wrong ID. (04:53) With Rift Lake African cichlids, it’s especially important because a lot of these species look very similar but can have different requirements and compatibility issues. To a random person, this seems random, but it’s not. Ashraf is a fishkeeper, and he’s put together an app-based solution for fellow fishkeepers. (05:12) He identified a use case in a niche he has experience in - as a founder might. Good job, Ashraf. So, this is his African Cichlid Fact Finder that he has decided to name: Know Your African. I take back what I said about the good job. Users fill in a common name of a fish That triggers Zapier to send a prompt to OpenAI asking it for the scientific name of the fish. (05:40) And then OpenAI returns that completion into the scientific name column, also via Zapier. Another two prompts: First, when a new scientific name is entered, Zapier asks OpenAI for the natural habitat of the fish based on that scientific name. Second, a prompt asking Dalle to create an image based on the scientific name. (06:01) The integration works, but there is a bigger problem. The facts are incorrect. This is what Ashraf wrote: Problem is OpenAI is almost always wrong. It returns wrong scientific names based on common names, and wrong habitats based on scientific names, so beginner fishkeepers using this app could be misled from the start. (06:22) A beginner who follows this advice could make mistakes that kill their fish. I was hoping as a safety measure we could use AI-generated pictures based on the scientific name but they’re way off! For example, row 13’s yellow lab’s photo is a completely different fish. This is what a yellow lab looks like. (06:39) But this is what Dalle gave me: If I came across this app, there’s just no way I’d trust it and recommend it to other fish keepers. Sorry, Ashraf. They all look the same to me. The only fish I know looks like this: Now, Ashraf used Google Sheets to simulate a frontend, but you could do the same with anything that can be an actual frontend and is part of Zapier’s 5000-app ecosystem. (07:02) Under their App Building section you’ll find all the usual no and low code app builders with pre-built integrations. That’s the integration part, we need to address those wrong answers. I appreciate that Ashraf didn’t just send in a prompt for something like a haiku about mountains and streams’ where there can’t be wrong answers. (07:20) His prompts have clear wrong or right answers - there’s one correct scientific name for a species. And he used the Davinci model - that’s what powers Chat GPT. OpenAI got it wrong - and of course it did. The model hasn’t been trained to be knowledgeable on fish or connected to real-time website scrapers. (07:39) Ashraf recognized the wrong answers because he knows better. But he’s right, a beginner could easily be misinformed. That applies to every other app relying on an untrained model. If you just want to validate an idea or get practice, sure, but long-term, your app can’t be giving out incorrect facts. (07:56) Anyways, that’s the Zapier way. The easier way. If you want to run a full blown app with that convenience, you’ll probably need to pay for it. Say you don’t want to pay - there’s always the low code way. There you go. You get a working Open AI integration with just a bit more work, but you don’t have to pay Zapier a dime. (08:52) Hey guys, you’ve just seen that OpenAI has made it pretty straightforward to integrate its AI models into your app. Would really appreciate it if you would subscribe and like the video. Leave a comment on which integration method you prefer. What comes next is a bit of a rant, but I promise it comes from a place of love. (09:10) A love of destruction! Kidding. What is a need you sense? Amongst real users. This is your use case. I think you - as in the viewers, You’ll be split into two groups. Those who want to include OpenAI or ChatGPT into an existing product, and Those who don’t have a product yet, but want to grab a slice of the AI pie. (09:32) If you fall into the first category, it’s pretty straightforward Find out how you want to improve your app Get your API key Identify the model you need Write that prompt using the OpenAI API playground where you can quickly switch between models and factors like how creative you want answers to be Test, test, test, then Release when ready For those who want to build an app that revolves around an AI model, I’m sorry. (09:58) Uncle Adrian needs to nag a little. If you go onto TheresAnAIForThat or any other AI tool aggregator, you’ll find thousands of apps. I feel like most of them are just clones of each other Before you build your AI-powered app, ask yourself. Do we really need another The answer is…yes! Users don’t want a monopoly. (10:22) But we need options. Not clones. I want to show you how OpenAI has made basic AI app development so much easier. Look at this. These are all official premade prompts that send requests to OpenAI. 48 different prompts, a diverse range of use cases. I see it as OpenAI’s way of saying ‘Here are some basic ways you can use our AI model”. (10:46) But those basic ways have been turned into basic apps with clone after clone running around calling themselves AI powered. But how many are just a glorified OpenAI prompt with an extra screen? Reminds me of this tweet by a legit SaaS player, Jon Yongfook. He’s expressing how I feel perfectly. I don’t want to discourage you from trying. (11:06) Go ahead and paste the playground prompts into your app builder or Zapier for practice. I just want to encourage you to think of a more rewarding way. A real AI product offers something more than a basic prompt. You can modify the prompt, and prompt engineering is its own field. People pay for that. (11:23) But ultimately, and this is just my opinion, But ultimately, I think AI apps that truly make an impact are those that take the time and effort to fine tune their own models. In fact, if we go back to that tweet, that’s exactly what the first reply is recommending. As Ashraf found - base Open AI models are inexperienced, and only ‘work’ if you have no idea what good answers look like. (11:47) I came across Visual Eyes some time ago. It's an AI powered heatmap tool. It scans your designs and predicts what draws users’ attention. Loceye, the company behind Visual Eyes, conducted large-scale eye-tracking studies with real people for over 100 websites to train VisualEyes' prediction capabilities. (12:06) Can’t imagine it was easy. But for their troubles, they have a unique product with thousands of users. As far as I can tell they didn’t use OpenAI, but that’s not the point. Point is these models need to be trained to maximize their value. And OpenAI is happy to help you fine tune a model. They charge a fee to fine tune any one of their gpt3 models to get better at specific tasks and subjects. (12:28) They even help transform your data into optimal formats to train the model. You’ll probably have to include someone with technical knowledge for this part because you need to use python and OpenAI’s Command Line Interface. But that’s how you build your VisualEyes. And just like with VisualEyes, no one else can use the model you've trained. (12:46) It’s your model. I go back to the 194 writing apps on TheresAnAIForThat. A question I ask myself when I use some of these apps: If I go straight to ChatGPT, would I get the same answer? If yes, I don’t think there’s a product here. It’s a data supply chain you’re asking people to use your way - the longer way - for no reason. (13:04) In app development, it’s so common to use the latest technology just for the hype. If you’re planning to use OpenAI before you have a use case, you’re skipping a foundational step in app development. An audience. A problem worth solving. Find those first. If you still need OpenAI, then look at implementation. (13:22) And as you can see, that’s just an API key away. I’ll install it for you for a price But first, you need to come with an idea worth the effort and expense. OpenAI is not going anywhere, and building a generic AI-powered app will get you nowhere. Watch this video on how to come up with app ideas customers actually want. (13:42) Then combine your knowledge of OpenAI models and app idea research and you’ll come up with something that addresses a real need. That’s the plan, anyway. Please subscribe and give this rant a like. I showed you how to use the API first right? You get what you want. I get what I want. Now let’s see how we can find you a million dollar app idea."
- Assistant
Title: How to Build AI-Powered Apps with OpenAI & ChatGPT (That Don’t Suck)
Transcript:
(00:00)
Introduction to OpenAI Integration
The integration itself isn't that complicated. The way I see it, if you don't have technical knowledge, you can do it with Zapier, or you can do it with a bit of coding. I'll show you. The coding way is trickier, but it has some advantages. What I really want to talk about is: if anybody can do it, should you? I promise, no nagging.
(00:20)
Oh wait, I apologize, there will be nagging.
First, let's give you what you came here for: how to build apps that revolve around OpenAI, or how to integrate OpenAI into an existing app.
Second, we'll look at how OpenAI has made AI app development really easy, and why that's not necessarily a good thing for would-be founders.
(00:41)
Understanding OpenAI Models vs. ChatGPT
Just to be clear, when people say "integrate ChatGPT," sometimes what they really mean is "integrate an OpenAI model." There are a few, and ChatGPT is an example of an app built with an OpenAI model. This one just happens to specialize in natural language production and surface-level domain expertise.
(01:05)
Emphasis on surface-level—more on that later.
If you want to create an app powered by an OpenAI model, you need to know:
- Your use case
- Available OpenAI models
- Their specializations
- Cost—they're not prohibitively expensive, but they're not free
Ahh... You've been using ChatGPT and thought it was all free? Nope! It's a free sample.
(01:26)
You want the real deal? It's going to cost you. Here's a list of available models together with their cost:
- GPT-4: GPT-3.5 on steroids
- GPT-3.5: Understand and generate natural language/code
- DALL·E: Generate and edit images
- Whisper: Convert audio to text
- Embeddings: Convert text to numerical form
I might have missed a few—there are a lot of them—but basically, you've got:
- Language models
- Image models
- Audio models
All of them have their place.
(01:55)
I know I've said it, but it's important and bears repeating: it comes down to your requirements. If you don't know, two options:
- Look at the models and try to find use cases that fit.
- Look for a use case, then find a model that fits.
These are the steps to integrate an OpenAI model into your app:
- Get a unique OpenAI API key by signing up with OpenAI.
(02:21)
- Construct a prompt and a frontend for users to interact with. Include specific instructions and necessary context. Normally, this means static instructions accompanied with dynamic user input.
Sounds unnecessarily complicated? Basically, you create a template, and your user fills in the blanks.
Super simple example: two text boxes on a screen.
(02:43)
One text box contains the prompt or static instructions, and the other is for user input, which changes with every request.
Here's an example of an input and output area in Google Sheets connected to OpenAI via Zapier.
Here's an example of a very basic app interface built from Locofy's training template. The input in the first box is part of the prompt sent to OpenAI.
(03:07)
OpenAI creates a completion. The second text box receives that completion.
- Evaluate the response: Was it what your users would have wanted? Adjust the prompt and test until you get accurate and reliable results.
Steps 1 and 3 are pretty straightforward. Step 2 is the tricky one. And there are two ways to do this if you're someone without coding knowledge.
(03:27)
Method 1: Using Zapier
The first and easier way is through Zapier. Zapier is a SaaS that simplifies data transfer between other apps. It's got common data transfer APIs built-in—you just arrange what you want.
It reminds me of this game called Rocket Mania. You've got a match on one end and a rocket on the other, and you arrange fuse blocks to establish the connection.
(03:50)
Same with Zapier. You don't have to build anything, just arrange what's already there and let the data flow.
Good news: Zapier has prebuilt connections with thousands of apps.
I got my writer to try this method out, and he has zero coding knowledge. Like, if I had to rate his coding out of ten, it would be "sucks balls out of ten."
He wrote that, by the way. I saw no reason to disagree.
(04:08)
He used Google Sheets as his frontend... and let's be charitable and give him that. If you put aside how it looks, it's an app. And his app is for people to get information on African cichlids.
What the hell is that?
African cichlids are a group of freshwater fish from the African continent. They're really colorful fish and popular aquarium pets.
(04:31)
I'm not a fish guy, so I asked Ashraf to explain his use case, and this is what he wrote:
"In fishkeeping, fishes can have many common names, which leads to confusion among beginners and even intermediates. However, there can only be one current scientific name, so fishkeepers prefer using that when doing research to reduce the risk of wrong identification.
With Rift Lake African cichlids, it's especially important because a lot of these species look very similar but can have different requirements and compatibility issues."
(04:53)
To a random person, this seems random, but it's not. Ashraf is a fishkeeper, and he's put together an app-based solution for fellow fishkeepers.
He identified a use case in a niche he has experience in—as a founder might. Good job, Ashraf.
(05:12)
So, this is his African Cichlid Fact Finder that he has decided to name: Know Your African.
I take back what I said about the good job.
Users fill in a common name of a fish. That triggers Zapier to send a prompt to OpenAI asking it for the scientific name of the fish.
(05:40)
And then OpenAI returns that completion into the scientific name column, also via Zapier.
Another two prompts:
- First, when a new scientific name is entered, Zapier asks OpenAI for the natural habitat of the fish based on that scientific name.
- Second, a prompt asking DALL·E to create an image based on the scientific name.
(06:01)
The integration works, but there is a bigger problem: the facts are incorrect. This is what Ashraf wrote:
"Problem is OpenAI is almost always wrong. It returns wrong scientific names based on common names, and wrong habitats based on scientific names, so beginner fishkeepers using this app could be misled from the start.
A beginner who follows this advice could make mistakes that kill their fish. I was hoping as a safety measure we could use AI-generated pictures based on the scientific name, but they’re way off! For example, row 13’s yellow lab’s photo is a completely different fish.
This is what a yellow lab looks like: [Image of yellow lab]
But this is what DALL·E gave me: [Image from DALL·E]"
(06:39)
If I came across this app, there’s just no way I’d trust it and recommend it to other fishkeepers.
Sorry, Ashraf. They all look the same to me. The only fish I know looks like this: [Image of different fish]
Now, Ashraf used Google Sheets to simulate a frontend, but you could do the same with anything that can be an actual frontend and is part of Zapier’s 5,000-app ecosystem.
(07:02)
Under their App Building section, you'll find all the usual no-code and low-code app builders with prebuilt integrations.
That's the integration part; we need to address those wrong answers.
I appreciate that Ashraf didn't just send in a prompt for something like a haiku about mountains and streams, where there can't be wrong answers.
(07:20)
His prompts have clear wrong or right answers—there's one correct scientific name for a species. And he used the Davinci model—that's what powers ChatGPT. OpenAI got it wrong—and of course it did. The model hasn't been trained to be knowledgeable on fish or connected to real-time website scrapers.
(07:39)
Ashraf recognized the wrong answers because he knows better. But he's right; a beginner could easily be misinformed. That applies to every other app relying on an untrained model. If you just want to validate an idea or get practice, sure, but long-term, your app can't be giving out incorrect facts.
(07:56)
Anyways, that's the Zapier way—the easier way. If you want to run a full-blown app with that convenience, you'll probably need to pay for it.
Say you don't want to pay—there's always the low-code way. There you go. You get a working OpenAI integration with just a bit more work, but you don't have to pay Zapier a dime.
(08:52)
Wrapping Up and Words of Caution
Hey guys, you've just seen that OpenAI has made it pretty straightforward to integrate its AI models into your app. Would really appreciate it if you would subscribe and like the video. Leave a comment on which integration method you prefer.
What comes next is a bit of a rant, but I promise it comes from a place of love.
(09:10)
A love of destruction! Kidding.
Evaluating the Need for Your AI-Powered App
What is a need you sense among real users? This is your use case.
I think you—as in the viewers—you'll be split into two groups:
- Those who want to include OpenAI or ChatGPT into an existing product
- Those who don't have a product yet but want to grab a slice of the AI pie
(09:32)
If you fall into the first category, it's pretty straightforward:
- Find out how you want to improve your app.
- Get your API key.
- Identify the model you need.
- Write that prompt using the OpenAI API playground, where you can quickly switch between models and factors like how creative you want answers to be.
- Test, test, test, then release when ready.
For those who want to build an app that revolves around an AI model, I'm sorry.
(09:58)
Uncle Adrian needs to nag a little.
If you go onto There's An AI For That or any other AI tool aggregator, you'll find thousands of apps. I feel like most of them are just clones of each other.
Before you build your AI-powered app, ask yourself: Do we really need another?
The answer is... yes! Users don't want a monopoly.
(10:22)
But we need options, not clones.
The Problem with Basic AI App Clones
I want to show you how OpenAI has made basic AI app development so much easier.
Look at this. These are all official pre-made prompts that send requests to OpenAI. Forty-eight different prompts, a diverse range of use cases. I see it as OpenAI’s way of saying, "Here are some basic ways you can use our AI model."
(10:46)
But those basic ways have been turned into basic apps, with clone after clone running around calling themselves AI-powered. But how many are just a glorified OpenAI prompt with an extra screen?
Reminds me of this tweet by a legit SaaS player, Jon Yongfook. He's expressing how I feel perfectly.
I don't want to discourage you from trying.
(11:06)
Go ahead and paste the playground prompts into your app builder or Zapier for practice.
I just want to encourage you to think of a more rewarding way.
A real AI product offers something more than a basic prompt. You can modify the prompt, and prompt engineering is its own field. People pay for that.
(11:23)
But ultimately—and this is just my opinion—but ultimately, I think AI apps that truly make an impact are those that take the time and effort to fine-tune their own models.
In fact, if we go back to that tweet, that's exactly what the first reply is recommending.
As Ashraf found—base OpenAI models are inexperienced and only "work" if you have no idea what good answers look like.
(11:47)
Fine-Tuning and Creating Unique AI Models
I came across VisualEyes some time ago. It's an AI-powered heatmap tool. It scans your designs and predicts what draws users' attention. Loceye, the company behind VisualEyes, conducted large-scale eye-tracking studies with real people for over 100 websites to train VisualEyes' prediction capabilities.
(12:06)
Can't imagine it was easy. But for their troubles, they have a unique product with thousands of users. As far as I can tell, they didn't use OpenAI, but that's not the point.
Point is, these models need to be trained to maximize their value. And OpenAI is happy to help you fine-tune a model. They charge a fee to fine-tune any one of their GPT-3 models to get better at specific tasks and subjects.
(12:28)
They even help transform your data into optimal formats to train the model. You'll probably have to include someone with technical knowledge for this part because you need to use Python and OpenAI’s Command Line Interface.
But that's how you build your VisualEyes. And just like with VisualEyes, no one else can use the model you've trained.
(12:46)
It's your model.
I go back to the 194 writing apps on There's An AI For That. A question I ask myself when I use some of these apps: If I go straight to ChatGPT, would I get the same answer? If yes, I don't think there's a product here. It's a data supply chain you're asking people to use your way—the longer way—for no reason.
(13:04)
Focusing on Real Needs and Problems
In app development, it's so common to use the latest technology just for the hype. If you're planning to use OpenAI before you have a use case, you're skipping a foundational step in app development: an audience, a problem worth solving.
Find those first. If you still need OpenAI, then look at implementation.
(13:22)
And as you can see, that's just an API key away.
I'll install it for you—for a price. But first, you need to come with an idea worth the effort and expense.
OpenAI is not going anywhere, and building a generic AI-powered app will get you nowhere.
Watch this video on how to come up with app ideas customers actually want.
(13:42)
Then combine your knowledge of OpenAI models and app idea research, and you'll come up with something that addresses a real need.
That's the plan, anyway.
Please subscribe and give this rant a like. I showed you how to use the API first, right? You get what you want. I get what I want.
Now let's see how we can find you a million-dollar app idea.