Open AI transcription

If you're not already building an AI-focused side hustle, you are behind the curve.
ChatGPT freaked programmers out in the same way the Benz motorcar freaked horses out 150 years ago.
It could be all hype, or it could be a paradigm shift into the artificial intelligence age.
Either way, if you listen carefully, you can hear a stampede of developers in the background getting in on the next technological gold rush.
But like I've said before, during a gold rush, sell shovels. If you want to make money. This video is my free shovel to you, where we'll take a realistic look at how to build AI-driven apps from both technical and business perspectives. There's a lot of VC money flowing into AI startups right now. We'll examine them to see how their tech is implemented and how they plan on making money. And you might be surprised at how easy it actually is to build an AI app when using the right tools. The ultimate shovel salesman is OpenAI, the creators of ChatGPT, which is based on an autoregressive language model called GPT-3, or technically 3.5. By now we all know how impressive it is, but what's crazy is GPT-4 is around the corner. GPT-3 is trained on an unprecedented 175 billion parameters, but that's absolutely dwarfed by GPT-4 with 100 trillion parameters. Maybe we've reached the peak of what is possible with deep learning, and GPT-4 is a nothing burger. OpenAI's CEO says you'll be disappointed, but maybe that's what he wants you to think because he knows the technological singularity is near, in which all of humanity is enslaved or exterminated. Either way, the only thing that really matters in life is that we make some mother effin money. One idea you might have is to build your own superior GPT model, but that's an idea I would describe as completely s**t. You're up against a company with the smartest people in the world that's received billions of dollars in funding from companies like Microsoft. If you can't beat them, join them. Their technology is available via a paid API, and just recently became available on Azure Cloud, and there's all kinds of use cases if you use your imagination. One example is Videotapit, which takes a video as an input and converts it into a blog post, by transcribing it and using AI to generate headers automatically. An app like this could be accomplished by an independent developer or small team. You could use OpenAI Whisper to do the transcription, GPT to generate the headers, and you could even throw in something like Dolly to generate unique images for the blog post. Now I realize 5 seconds ago I said you can't compete with OpenAI, but don't listen to haters like me. GPT is not the only game in town. You have DeepMind from Google, with other language models like Chinchilla, and AlphaCo that's been out there winning coding competitions, and there's also open source language models like Bloom. If you have access to a lot of high quality data, it is possible you could train your own predictive model, then offer it as a service. The process of turning raw data into an actual deployed machine learning model is called MLOps. Tools on Google Cloud like Vertex AI handle every step of the process. They ingest and process data, they train and evaluate models, and then deploy them to serverless infrastructure where you can actually use them in an app. In addition, there are tools like Hugging Face that provides MLOps as a service, but they provide thousands of open source pre-trained models that you can use as a starting point. Most importantly, you'll want to get really good at some kind of specialized use case. An example that comes to mind is Play.ht, which can create custom AI generated voices. They have an entire fake podcast between Joe Rogan and Steve Jobs, and for all you know, my voice could be entirely AI generated right now. You can clone your own voice. It is crazy good. Uh, it's not. My name is Jeff. But I did look into it, and it costs tens of thousands of dollars to get your voice cloned, which is understandable because there's very few companies that can actually offer this service at a high level, at least in today's world. Another crazy example is Beat Oven, which I'm using right now to generate this inappropriately tense background music. It combines generative AI with real music to create unique music. Or how about Resume Worded, a tool that can analyze your resume and LinkedIn profile. If you never learned to read good or write good, AI can now do all the heavy lifting. And I guarantee you there are developers right now building AI apps that will help kids cheat on their homework. At the same time, other developers are building tools to prevent AI cheating by using AI. There's a ton of different angles to make money here. No matter what you decide to build, it's a good idea to build in public, which can sometimes make your business go viral. That's exactly what happened with RestorePhotos.io. It's a simple Next.js app that was launched a few weeks ago and already has over 40,000 users. What I love about it is that it solves one simple problem. You upload a grainy photo, then it uses AI to restore it to a high resolution photo. The code is all open source, and as you can see here, it uses an API called Replicate to perform the image transformation with an open source machine learning model. In this case, the model is GFPGAN, but there are many other models to choose from. The key here is to find a problem, then create a no-nonsense, highly focused solution for it. Then finally, do some kind of TikTok dance while using it to make it go viral. Another potential way to make money is to slap the label of artificial intelligence on an existing business model that already works. We've already seen many companies do that in recent years, like McDonald's going fully automated, which is great because it means no unwanted reproductive fluids in your Big Mac, or like Upstart, an AI lending platform that uses machine learning, or a fancy way of doing statistics, to lend money on more factors than just the credit score. Its IPO investors are currently holding the bag for a 90% loss, but its early investors got rich. You could build an app like this yourself without even using deep learning. You just need like 10 lines of Python code to run a regression on the input data, then boom, you're now an AI lending platform, assuming you also own your own bank. The takeaway here is that you can use the artificial intelligence label to make your business look disruptive, even if it's running a grip that's been around forever. Now if you really want to make a lot of money, don't look around at other cheap poor people like yourself. Instead, sell to enterprise and government. Finish this sentence, blank.ai provides a tool that will save your company 70% per year in blank costs. Congratulations, you just wrote a pitch for the next AI unicorn startup. The blank might be something like shipping logistics, talent recruitment, or business presentations, as is the case with present.ai. This startup will take a company's branding and marketing assets, then automatically create tons of slide templates for business presentations. I have no idea what their tech looks like under the hood, but it looks like a use case for generative image models, or tools that can take text as a prompt and return images as a result. One of the most impressive tools in this space is Midjourney. You can use the beta by going to their discord and ask for something like a business presentation for my oil and gas company. The result you get back looks like something produced by a talented human. Generative AI is so hot right now, but there's just as much if not more opportunity in natural language processing. Another huge cost for enterprise is customer support. ChatGPT has made it clear that customer support jobs will largely go to AI in the coming years. However, the average company has no idea how to integrate this technology into their business. Moveworks is just one of the companies that is bridging this gap. They build AI chatbots that integrate with your entire tech stack that can solve problems faster and cheaper than a human could ever dream of. There are tools like AWS Lex and Google Dialogflow that you can use to build chatbots right now. They can handle business logic on the back end with webhooks, and can be integrated with GPT-3 to provide unique human-like interactions. As you can see, there are going to be tons of opportunities in the next few years for ambitious and clever developers. However, there is a dark side to all of this. The technology is moving so fast that an idea that might look like a guaranteed success today turns out to be completely obsolete a year from now. In addition, the sad reality is that big tech is going to absolutely dominate this space. They have all the data and computing power required to really push this technology forward. It's really hard for me to compete with my Intel Pentium processor from my mom's basement. Even if you do manage to build a game changer, Zuckerberg is just going to steal it anyway. Another dark side to AI is that this technology will likely be harnessed by scammers. It's actually pretty scary. Someone could clone my voice based on all my YouTube videos, then call my grandma and tell her I got arrested in Mexico and that she needs to transfer some bitcoin over to get me out. Not cool to scare my grandma like that, man. Also, at one point or another, everybody has wired money over to their long-lost Nigerian prince relative, only to be disappointed when that big inheritance never comes through. Scams like this are only going to get better. We can scale them up by using tools like ChatGPT to automatically format letters in proper English. Um, well actually, upon testing, I guess scam emails are not allowed. In addition, you could even target a specific person and combine their face with some real Nigerian prince so the target actually sees a long-lost relative with resemblance to their own face. We're entering a brave new world where it's going to be very difficult to distinguish fakery from reality. Deepfakes have been around for a while, but they're only going to get better. Big companies that have all your data know everything about you. They know what you read, how you write, how you look, and how you act, and they've been using that data to serve you advertisements. But nowadays with generative AI, they quite literally have the ability to become you in the digital sense. Sounds creepy, but the good news is that we can use this technology to live forever. Even if I get smashed by an elephant or fall into a volcano, I will continue making these YouTube videos because the artificial me will be indistinguishable, nay superior to the real me, and won't make silly typos. Not only will I continue working, but I will continue to raise my family in the metaverse. And not just my kids, but their kids, and then their kids. Because I will be the apex alpha artificial grandpa, the first of its kind. They could even upload me to a lifelike robotic doll so my wife and kids never even know I died. That's actually a great business idea. Create a company that clones a person's persona, uploads it to a doll that is automatically deployed when that person's organic form terminates. That may sound crazy, but it's within the realm of possibility in our lifetimes. Thanks for watching, and I will see you forever and ever until the end of time.