AI apps instead of prompting: efficient solutions for specific tasks

Chatbots are often misused as universal assistants. This involves entering a complex prompt or manually prompting in several steps. Automation and reliability are different. With the help of specific AI apps for specific tasks, the degree of reliability and automation can be significantly increased.

Introduction

Does your AI system ask you "What do you want to do?" ?

Probably yes, because many universal systems act as if you can solve any kind of problem with their help. To a certain extent, this is true. But only to a very limited extent.

Below you will find a video on the topic Secure & Optimized AI with AI Apps The video runs locally on a German Dr. GDPR server and uses adaptive streaming (adaptation to your bandwidth).

Unreliable solutions can work for one-person users who check every result on their own responsibility because they know that they themselves are liable for it. These individual users also like to break down a complex task into sub-steps. Not because they feel like it, but because the AI system (ChatGPT, Gemini, etc.), which is immature for the specific task, forces them to do so.

What if an AI solution offered users exactly the options that the AI system is really capable of?

The picture shows concretely designed AI apps:

• AI wiki: Users can query company knowledge. It is not world knowledge from the USA or China that is surveyed, but what your company has defined as correct and relevant
• Mini-apps: Very small, self-contained applications, for example:
• Summarize text: without prompting
• Translating text: without prompting
• Writing articles: without prompting

Below are more complex apps that cannot be created with ChatGPT (if data security is not an issue).

Let's take a closer look at the apps. The picture also shows "General chat". This can actually "only" be used to query world knowledge. No company actually needs its own chat, as ChatGPT already exists. You "only" pay with your data. As long as no employee accidentally (or carelessly) enters sensitive data into ChatGPT, everything is fine. However, if sensitive data is entered into the ChatGPT prompt, confidentiality is no longer an issue.

What is an AI app?

An AI app is optimized for a specific application, i.e. for a specific problem!

An AI app is typically based on a powerful open source AI model. In addition, data preparation is always carried out before and after the AI process to increase reliability. You will not find this very important data preparation with ChatGPT!

An AI app is best run in Germany. This means a purely German provider, without an Irish subsidiary or American parent company. You can provide the hardware yourself or rent it.

The costs for an AI app are manageable. We will go into this later.

In the following, we differentiate between fully-fledged AI apps and mini-apps for a better understanding of the possibilities offered by AI.

Fully-fledged AI apps

A fully-fledged AI app is a monster. This type of app is more complex than the mini-apps shown below.

A good example for a full-fledged app is a Company-Wiki. With this, thousands of own documents can be queried. The process is as follows:

1. All relevant documents from your company are read into the AI system once or periodically. File formats could be: PDF, Office files, text files, images
2. All documents are automatically prepared by your AI and made more searchable. Forget ChatGPT
3. You ask your AI a question
4. Your AI responds on the basis of your own tens of thousands of documents

Does this also work with ChatGPT?

No, because

• ChatGPT does not read thousands of documents for you;
• ChatGPT does not prepare your documents the way they need to be;
• If you were to use the ChatGPT API, you would have to pay per call. The costs depend on the scope of all documents and are incurred anew each time;
• We're not even talking about data security here.

Another good example for a full-fledged AI-app is the Simulation of Personas for Market Research.

The idea is simple: the user of the personas AI defines different buyer types, for example the price-conscious father, the environmentally conscious single woman or the digital shopper. All personas can be assigned social characteristics and behaviors so that the simulation is as realistic as possible.

Now these personas can be questioned at any time and as often as desired. By questioning individual personas, advertising texts and packaging, or even promotional actions, can be optimized. The POS marketing (Point of Sale) can thus be generally optimized, without having to search for real study participants and torment them with money at a specific time.

All of this can be done without a cumbersome prompt. The only "prompt" that is entered is the question (or request) to the personas AI. However, this question or request is not a prompt, but a question or input as we know it from forms.

The AI can distinguish which knowledge it needs to tap into in order to answer the question you ask as a market researcher:

• Personas character;
• POS marketing knowledge that is imported into the system once and can be updated as required;
• 100,000 promotions that you have accumulated in your database over the years.

It has never been easier to obtain realistic statements from virtual shoppers. Advertising slogans can be found or optimized with much less effort and more easily (energy savings!).

The list of (real-life) examples could go on and on. Further examples of fully-fledged AI apps will be mentioned shortly:

• Accounting AI: Evaluation of invoice positions and creation of reports with the result of the evaluation
• Item reconciliation