What are custom GPTs?
Custom GPTs or AI assistants are customized and individual AI systems that are specially tailored to specific needs, requirements or industries. In contrast to general AI applications, such as the well-known ChatGPT, which cover a wide range of topics and questions, custom GPTs are trained on specific domains, data sources or task areas. As a result, they offer personalized answers and are better able to respond to context-specific questions and challenges and complete tasks for their users.
For example, a CustomGPT could be developed for a company, a brand or a specialist field and integrate the knowledge, processes and vocabulary of this company in order to subsequently write suitable marketing texts for the company. Employees can use the Custom GPT on their own or share it with the team. This ensures that the tone of voice remains consistent and at the same time significantly increases employee productivity by supporting the AI model. You can find out more about custom GPTs in my article on implementing custom GPTs.
What examples of custom GPTs are there?
Custom GPTs have become known primarily through the GPT Store from OpenAI. Every Chat GPT user can access the GPT Store via the navigation on the right-hand side using the “Explore GPTs” button and then use specific GPTs developed by others.
Here is a small selection of my GPTs
1. branding partner
This GPT creates guidelines for communication based on links and documents that you make available to it. To the GPT.
2. AI Language Tandem
You can learn languages with this GPT. To the GPT.
3 Frida – For relatives of people with dementia
This GPT plays a role play with you so that you can learn how to communicate with people affected by dementia. To the GPT.
Is personal data protected with Custom GPTs?
The use of the OpenAI custom GPTs shown above is subject to the same data protection guidelines as the use of the general ChatGPT. This means that all input data is always processed further by OpenAI. This means that the data entered in the input field of the custom GPT is not protected and users can only use this type of AI assistant for applications without sensitive data.
It must also be ensured that no confidential data is passed on to OpenAI when the custom GPT is created. Custom GPTs are personalized by specific prompt instructions and by uploading your own knowledge sources. However, this knowledge is not protected, but is also accessible to OpenAI.
How do I create AI assistants that take data protection into account?
The advantages of individual AI assistants are enormous. Thanks to the ability to customize the AI Assistant to your own needs by giving it documents and prompts as behavioral instructions, AI Assistants are a powerful tool. However, anyone who is concerned about their personal data must choose alternatives to the custom GPTs from OpenAI.
One possible alternative is Peak Privacy. In addition to the general AI chat, such as ChatGPT, Peak Privacy also offers the option of creating your own AI assistants, known as “constructors”.
How can I implement AI Assistants in compliance with data protection regulations?
A data protection-safe option for implementing AI assistants, analogous to custom GPTs, is possible with the AI assistant from Peak Privacy, among others. In addition to the classic prompting and uploading of files, the tool offers significantly more language models than OpenAI, does not store or use any personal data, offers on-premise hosting or clouds in Germany and Switzerland and provides even more customization options than OpenAI.
The implementation of data protection-compliant AI assistants in detail
Below are the most important steps in the implementation of AI Assistants using the example of Peak Privacy. AI assistants are referred to here as constructors.
General settings
As can be seen in the image, the creation of the AI assistant or constructor starts with the configuration of a few general settings, such as name, description and image.
Select the language model
With Peak Privacy, users can choose from a variety of language models. Many of them are open source and thus ensure that no personal data is passed on. So if you want to implement a data protection-compliant AI assistant without personal data going to companies such as OpenAI, choose llama-3, for example.
The system description – Prompting
The system description is the heart of the AI Assistant or designer, so to speak. This is where you define how the individual AI assistant should behave, what tasks it should perform and what questions it should ask its users. You can also enter your own brand tonality here. The “Branding GPT” can again be used to create tonality guidelines. This custom GPT can be accessed here.
Important: When creating the system description, the same prompt rules apply as for ChatGPT. More on this in the prompt instructions article.
Further configurations
In the next step, users have the opportunity to personalize their individual AI Assistant to a much greater extent than is possible with OpenAI’s custom GPTs by means of additional setting options. For many users, especially those who are already familiar with the use of language models, this is a great advantage. For others, however, the multitude of customization options can be overwhelming. If you do not want to make any further adjustments or lack the necessary expertise, you can use the AI Assistant with the default values and simply apply the personalized system prompt.
Testing and use
As soon as all settings have been made, the AI Assistant can be used and tested. In most cases, individual adjustments are made to the system settings after the first few uses and then the individual AI Assistant can be used. You have the option of using the designer alone or sharing it with the team.
Incidentally, the AI Assistant can also be used by voice input.
Frequently asked questions about data protection at AI Assistants?
Who has access to the data entered?
When using secure applications such as Peak privacy or SwissGPT, the data entered is not used any further and third parties have no access.
When using OpenAI or Google Gemini, the input data is further used by the respective companies for training purposes.
What happens to the documents used to train the AI Assistants?
When using secure applications such as Peak privacy or SwissGPT, the training documents are only available to the creator of the AI Assistant.
What does the data protection-safe implementation of AI assistants cost?
The exact pricing depends on the scope and use of the AI Assistant. As a rule, the license costs start at EUR 50 – 100. Added to this is the effort required to implement or create the system settings. Find out more in the GPT offers.
How long does it take to implement privacy-safe AI assistants?
In principle, privacy-safe AI assistants, such as those designed by Peak Privacy, can be implemented in just a few hours or days.
How are AI Assistants shared within the team?
AI Assistants from Peak Privacy can be shared with other team members via a sharing function. It should be noted that only team members have access to the AI Assistant. This is where the function differs from OpenAI’s GPTs, which can also be shared publicly.
My offers for the implementation of CustomGPTs and AI assistants
Jump start
For individuals
In a one-to-one coaching session (video call), I will give you the most important tips and tricks for implementing a CustomGPT. You will benefit from my experience and, with a little patience and skill, you can then create your own CustomGPT or AI assistant.
Joint implementation
For start-ups, NGOs and SMEs
We will implement the CustomGPT together and you will then be able to make adjustments yourself. As a rule, implementation is a process consisting of various coordination calls and joint and separate implementation sessions. Depending on the team, the AI assistant will be ready for use in 1-2 weeks.
Jump start
For individuals
In a one-to-one coaching session (video call), I will give you the most important tips and tricks for implementing a CustomGPT. You will benefit from my experience and, with a little patience and skill, you can then create your own CustomGPT or AI assistant.
This article is also available as a podcast episode
Attention! The podcast was created entirely by my AI-Assistant based on my contribution – no guarantee for incorrect content.
