-
OpenAI catered to business concerns about GenAI with the launch of ChatGPT Enterprise
-
But the product still doesn't address the possibility for hallucinations, Gartner's Jim Hare said
-
The release could spark the launch of more vertical-specific AI products but companies should be wary of model diversity becoming too complex
ChatGPT put on a shiny new suit at the end of August, unveiling its upgraded service tier tailored to enterprises. According to Gartner VP and Analyst Jim Hare, the tool still presents data privacy and security challenges, but it may initiate a trend towards more segmented, industry-specific GPT applications.
OpenAI opened up ChatGPT to the public late last year, and the generative AI (GenAI) swiftly became one of the most quickly popularized internet apps in history — surpassing 100 million users in just two months. The company officially announced the launch of ChatGPT Enterprise, offering “enterprise-grade security and privacy” and upgraded priority access to its bellwether AI engine, GPT-4.
Hare explained to Silverlinings that enterprises have been “pretty jazzed about GenAI” and curious about GPT chatbots as a tool. But when they began hearing of issues and risks associated with data privacy and information accuracy, businesses were then pocketing the tool as something “great to play with” but unable to really use fully for proper enterprise purposes.
“So they were looking for alternatives,” he continued, prowling for other powerful AI models from Microsoft, Amazon and Google. Microsoft in particular has set itself up for success in the chatbot sandbox — investing billions into OpenAI, since ChatGPT runs on its Azure platform. Microsoft even had its own go at a chatbot-for-business service, announcing Bing Chat Enterprise earlier this summer.
But with this latest firm-focused version from OpenAI, the tool is addressing primary problems around data privacy — with OpenAI explicitly saying they will not use any prompt information as part of its training — and having encrypted, SOC 2-compliant information.
That doesn’t mean there aren’t still concerns. Among them: the possibility of hallucinations when the service “will only be as good as the information you give it,” Hare commented.
ChatGPT: a ‘productivity assistant,’ not a cure-all
Using ChatGPT Enterprise will require industry and company context for it to provide useful information. “The question then becomes, well, how much data can I give it? And if I want to give this thing access to my enterprise data, how do I make sure I do it in a controlled way, with role-based permissioning?” Hare said, adding that it that remains unclear, and potentially brings up another issue of sharing data across multiple clouds.
And much like a GPT model (only as good as the information it’s fed), the training of employees on how to effectively and safely use chatbots will continue to be vital — especially when human error charts as the highest cause for cloud security breaches.
Indeed, the need to bring employees along for the ride spans AI offerings from different vendors. During a customer roundtable for media at Google Cloud Next last week, Deutsche Bank's Chief Strategy Officer for Technology, Data and Innovation Christoph Rabenseifner said "the most important challenge we see here...is we have to train our people that they understand how these models work and what inputs are provided to the models and outputs come from them." He added Deutsche Bank has undertaken a program to train not just its technology staff but all of its employees on how its AI tools work.
But even with the early adoption challenges of OpenAI’s latest tool, Hare sees it propelling a trend that may reveal a “move away from large mega models to more domain-specific models.”
Be it the medical world or the legal field, tools that are specifically trained on industry-relevant data points may provide the needed “horsepower” and “contribute to the growth of GenAI in the enterprise.” However, he noted the model management may become more complex if they become too segregated.
Currently ChatGPT, even with its sleek new business suit, is only “a productivity assistant, versus something that is going to always give you accurate answers or responses. It’s not geared to do that right now,” Hare said. That means while it may be helpful to businesses in some specific use cases, in areas where accuracy really matters — such as customer support — the tool is likely to still only serve as a kind of polishing “wrapper” to the candy provided by people.
Editor Diana Goovaerts contributed Deutsche Bank's comments from Google Cloud Next to this story.