Featuring:
Rowan Curran, Analyst
Show Notes:
Every once in a while, adoption of an emerging technology explodes, driving business and technology leaders to assess its impact on the enterprise. In 2023, it’s generative AI (thanks ChatGPT). In this episode, Analyst Rowan Curran provides his perspective on the latest developments in the generative AI technology space and dispels some common misconceptions along the way.
Curran starts by explaining what generative AI actually is (models that underly certain applications) and, perhaps more importantly, what it isn’t (any single tool or application). For example, he explains that ChatGPT is an application built on top of generative AI technology, but “it is not generative AI in and of itself,” as some headlines may lead you to believe. To help listeners further understand the technology, Curran describes the pros and cons of the various types of models that power it including generative adversarial networks, diffusion models, and most recently, multimodal large language models.
But the big question on business leaders’ minds is: How will the powerful generative AI technology change business? Throughout the episode, Curran provides a variety of enterprise use cases for generative AI from image and text generation to chatbots to development. When asked what he sees as the next big step for the rapidly evolving technology, Curran says the ability to run generative AI models on a local machine or even on a smartphone without latency will change how the models are implemented, pointing to Meta’s recent LLaMA release as an early step in this direction. That step could lead to “personally trained large language models” that enable your smartphone assistant to be driven by a large language model that is fine-tuned based on the questions you asked it over the years.
Later in the episode, the discussion turns to the risks associated with overdemocratization of AI through the new tools. And there are plenty. Curran says the inherent variability that comes from large language models (the same prompt generating different results at different times) must be taken into account by technology leaders considering implementing these tools. For example, allowing your firm’s customer-facing chatbot to make up its own responses with no human oversight still brings a very high risk of brand or reputation damage. He adds that knowing what data set the model you’re using was built on can help with testing and governance.
The episode closes with some words of warning to enterprise technology leaders. While generative AI has tremendous potential for positive change, it also offers tremendous potential for malicious actors. For example, Curran points out that generative AI technology can be used to not only create much more realistic phishing emails but also to very quickly create entire websites and online experiences, which could make the battle against hackers more challenging.
