Many people have already had the opportunity to experience OpenAI’s ChatGPT service, which allows users to interact with the system in the form of a conversational bot. In comparison to traditional bots, the quality of ChatGPT’s responses has exceeded expectations. The recently released ChatGPT APIs give enterprises more options to explore business cases. We have been frequently asked about the trends around ChatGPT and what they mean to the enterprise. There may be a revolution coming on the heels of generative AI — but tech leaders should maintain a realistic perspective and understand its potential and limitations.

What Is ChatGPT?

Over the past couple months, ChatGPT has become nearly synonymous with “advanced AI” in the public mind. ChatGPT itself is just one specific application of the underlying models that have been trained by OpenAI, and so teasing out the application from the underlying technology is important to tech leaders looking to apply it themselves. When we look from the technology, application, and strategic levels, respectively, ChatGPT is:

  • A new model with supporting mechanisms. As described by OpenAI, ChatGPT uses a recent version of its GPT series of models, with this version officially being called “GPT-3.5-turbo.” The GPT-2 and GPT-3 series of models are available from both OpenAI and Microsoft as API services, and the GPT-2 model can be downloaded and built upon directly from Hugging Face. ChatGPT also includes further layers of processing and intelligence, driven by functionalities such as the reinforcement learning from human feedback.
  • A powerful chatbot application. The massive hype storm surrounding ChatGPT can’t be attributed to its technology and models alone — they’d been available for over a couple of months. So why did we see the surge in attention? Because ChatGPT delivers all of these capabilities into a compelling application experience. As tech leaders rush to understand how they should respond to the excitement around generative AI, they should keep this at the front of their mind. Even with extremely powerful technologies, such as large language models (LLMs), what will drive adoption success and practical usage is using them to build compelling applications that solve targeted business and user problems.
  • Proof of the evolving “AI 2.0” trend. The emergence of ChatGPT is a demonstration of the trend toward AI 2.0. Forrester defines AI 2.0 with three attributes: creative, generalizable, and pervasive. The recent advancement of ChatGPT not only shows exceptional ability in terms of text content creation but also reflects the versatility of scenarios that generalizable models can enhance. Companies must provide AI products and services based on AI 2.0 characteristics and at the same time meet the buyer’s business needs. After the service launch, ChatGPT has attracted great attention from the market and lit a fire underneath tech companies large and small, driving them to respond. This has resulted in a plethora of companies announcing that they will launch ChatGPT-like or OpenAI-driven products and services. While few of them are announcing publicly accessible LLM-driven chatbots, companies such as Google and Microsoft are now in a competition to deliver a compelling chat-driven search experience.

Can ChatGPT Really Do Everything?

No, it can’t. ChatGPT offers impressive responses, but it also produces hallucinations and coherent nonsense. LLMs such as those underlying ChatGPT have significant strengths but also a variety of caveats. Before leveraging ChatGPT in production environments, companies need to think about:

  • Model limitations. OpenAI has noted some limitations about ChatGPT. Enterprises should be mindful of the issues that have been highlighted because they not only apply to ChatGPT but also to many other applications of LLMs. For example, the summarized outputs produced by ChatGPT may not always accurately reflect the source data; the data used for model training is based on historical information (up to 2021), which restricts it in offering feedback for current trends. Additionally, ChatGPT has been known to generate outputs that are influenced by social biases. Companies need to evaluate how these issues may affect current product or service offerings before embedding ChatGPT.
  • Data compliance. The deployment of AI algorithms must consider the compliance requirements. For example, the deployment of LLM model services in China needs to comply with the data security law and regulations on algorithm recommendations. Although OpenAI claims that it will not use data submitted by customers and offers dedicated instances, companies still need to confirm with their legal and compliance teams before launching new products or services through third-party APIs or new models.

If you would like to discuss topics regarding generative AI, large language models, or foundation models, please schedule an inquiry (inquiry@forrester.com) or submit a briefing request.