When ChatGPT came out, suddenly, the excitement of AI became contagious. More and more ideas and possibilities came about on how to use AI. It has been a long time since we felt this way with tech. The last time I saw this level of excitement was when the smartphone existed. 

AI has increased the interest of people with little tech background and seeking to take their idea off the ground. If you are not technical, you’re seeking ways to get started on your own. I know what the excitement feels like in the early stages. You are pumped with energy and motivated to really hit the ground running. 

Without any technical background, you can easily want to default to the similarly available options for web development or app development. These no-code solutions allow you to drag and drop what you want, and then an app is made for you without you needing to learn how to code.

You need to remember that no code came about when the industry was mature. The options for building websites were tested, and the end results were clear after many repetitions and a well-defined framework for developing good codes. 

However, now you are entering into a period as an early adopter of new technology. Where technical infrastructures for AI development are not 100% well defined.  No-code will also be in the same position if you seek a no-code to build custom AI models for unique purposes. So, there will be limitations if you are looking for no-code options specifically for AI model development. But this is not to say that tomorrow someone develops this exact solution.

A few things need to happen for a no-code AI custom model-making application to exist. 

1. There needs to be a way that the no code can build a custom model.
   1. It will need to evaluate using your own data or other data, which are labeled or not. 
   2. It will then need to train on the data and split it into three groups: training, evaluation, and testing. How it decides to split the data is dependent on the problem.
   3. After that, it will need to review the training results to see if the model is trained well, has no bias, or needs more work. 
   4. It will automatically need to adjust some parameters and continuously train to get it right if it needs more work. 
   5. It will need to know when to stop training any further.  
   6. It will then need to save the model to be compatible with your device.

2. The model needs to fit automatically with the device or the cloud server.
   1. Once you have a model, you might want to use it on a website, a mobile app, or another special hardware device. The file format must be compatible with the device without changing its performance.
   2. The no-code solution would need to build out the functions that interact with the model and allow for inputs to send new data to the model for processing.
   3. The app or website needs to be deployed for others to use.

Using a generic AI model’s API with no-code

No-code can offer a great start to getting your product to market and allow you to test at a low cost. It allows you to bypass developers for some already customized plug-and-play pre-built AI model that allows you to build what you need in a couple of hours.

Most of the steps in part two above can be done with no-code programs like Bubble or FlutterFlow. You can use legacy no-code solutions for websites, apps, or any other popular front-facing app and connect an AI model using an API. These AI models will be generic. They will apply to everyday usage in finance, marketing, and other domains.

However, they do not cover everything you need to build custom solutions. The more niche, geographical, or domain expertise your idea requires, the less likely these pre-built models would work. The responses from the AI model will be incorrect, or it will start hallucinating, as they call it.

Other options could be low-code applications. These applications may allow you to piggyback an existing model for audio, speech, image, video, or text, either a document or a database, to fine-tune or make the model fit for your purpose. The model needs to have similar characteristics and data points that is learning on your new data for it to work.

Say that you want a model that had been previously trained on text documents, but the reading was done from left to right to comprehend and make sentiment analysis. But your new document data reflects another cultural norm where they read text from right to left. This could generate a different meaning and sentiment using the existing model. The existing model has been trained to understand and interpret text from left to right. So, adding your data document wouldn’t work. You wouldn’t be able to change the parameters in this low-code solution.

However, if your idea is about making a social impact, existing AI models will mostly like not work. I’m speaking more of people who desire to make a massive impact toward sustainable development goals that will impact people and the planet. Often these ideas will not have long-term success with no code solutions.

Sure, you can hire a developer to train on existing models that are used for ChatGPT. ChatGPT is good for generic problems. You can go one step further and train your own personal data using an existing ChatGPT model with the help of an AI developer. With your own custom model, you can wrap the ChatGPT API around a user interface using no code programs. Your custom API will operate in your app’s background. 

The key thing here is that you will need a developer to make that happen. Most of us are not making conversational applications. So there’s more work to do that will involve a developer. It requires that you build your model from scratch or leverage other models to assist in making your idea work.

Generic AI systems will lack the detailed knowledge and context needed for your idea.  I like comparing AI models like ChatGPT to entry-level positions or experienced professionals, whether it’s a marketing assistant, research coordinator, or some type of role with repetitive tasks. 

But when you are looking for a VP for international development in water sanitation, you’ve passed the level that a generic model can perform well. 

More importantly, when you are working in an area with varying contexts that are constantly changing and require years of knowledge and experience, you will need more than generic solutions.

Often when we have bigger ideas, like Chatbot Robots to connect with patients in hospitals or drones used to evaluate climate change effects, we’ve got to eliminate the option of using generic no-code solutions for now.

It requires more customization and more domain knowledge that your product needs to succeed. This means you will require more investment of resources to get to the same level of success as other generic products in the market.