Will ChatGPT be the new enterprise chatbot?

Last week, OpenAI publicly announced that there is now a version of ChatGPT that can be centrally deployed and administered by companies. In essence, this chatbot doesn't offer more than ChatGPT Plus. So why is this a game-changer? ChatGPT Enterprise does not share usage information with the GPT language model. This means that user inputs and the responses generated by ChatGPT are not used to train the language model.

Until now, there were two ways to avoid becoming another case like Samsung:

• Set individual user accounts to prevent their inputs from being used for training.
• Develop a custom chatbot that uses the language model and specifies that inputs shouldn't be used for training.

The first option isn't really viable for companies since their IT departments can't control individual ChatGPT user accounts and therefore can't ensure that the setting is applied. Plus, users lose their history. The second option requires development expertise and thus comes with associated development costs. Whether that's a disadvantage is currently unclear since the prices for ChatGPT Enterprise haven't been announced yet.

The subtle differences

Even though there aren't any truly new features, ChatGPT Enterprise is a step above both the free and Plus versions:

• There aren't any usage restrictions, unlike with ChatGPT Plus where only a certain number of prompts can be entered within a timeframe
• Prompts can be up to 32,000 tokens long
• It integrates a code interpreter
• Prompt templates can be shared

And it comes with enhanced speed on top of these. That's about its usage.

Secure and manageable

Good news for IT admins. The app can be integrated into a company's existing user management system. If that's not desired, there's a separate console for user management that supports bulk imports.

In terms of security, high standards like AES 256 and TLS 1.2+ are implemented for data encryption during storage and transport. On OpenAI's security portal, we learn that ChatGPT Enterprise is CCPA, GDPR, and SOC-2 compliant, among other things. However, it only provides an overview. Those wanting more details surprisingly need to sign an NDA with OpenAI and specify how ChatGPT Enterprise will be used. The good news is that approval takes less than 24 hours.

The data's journey

In Europe, when discussing security, the path data takes is also pertinent. According to OpenAI's privacy policy, their servers are located in the United States – and only there. This means that user data and any data generated during use are transferred to and stored there.

Unlike Microsoft, which offers the use of cloud services on servers in Europe or even Germany, OpenAI hasn't made such an offer and hasn't announced plans to.

More than just ChatGPT

What has been announced, however, is the ability to use ChatGPT Enterprise in conjunction with one's own data sources. Microsoft already offers this feature through its Azure AI Studio, but it doesn't meet all expectations. While sources with unstructured, text-based data integrate well, structured data doesn't. OpenAI's existing solution of working with Custom Functions has been more successful. Whether OpenAI can provide this advantage in a way that everyone can use it remains speculative.

Unknown availability

It's also speculative as to how long it will take for ChatGPT Enterprise to be ready for use in one's own company. So far, it's not possible to simply order ChatGPT Enterprise or even learn about its licensing costs. Instead, all inquiries go through OpenAI's sales team, which not only asks for contact information, location, and company size during the initial contact but also inquires about usage scenarios for ChatGPT Enterprise. Then, the waiting begins, as there's a high demand according to the manufacturer.

When the wait ends

We continue to wait, wanting all our questions answered. Especially whether ChatGPT Enterprise has the potential to become THE internal company chatbot. The best way to find out is by testing it ourselves and with our clients. Once we reach that point, an update will follow – that's a promise.