Not Every "AI" Use Case Requires The Latest and Greatest LLM or Multimodal Model
Frequently when I speak with investors I find they conflate Generative AI with other permutations of Artificial Intelligence. For example, AWS, Azure and GCP have made core AI services, including machine learning, available to corporate customers for approximately a decade. These advanced automation capabilities have been widely deployed for years.
When I hear investors say that companies are laying the groundwork for an “AI economy” I cringe. Companies have widely deployed “AI” for years across a number of use cases. Predictive Analytics (AI is nothing more than predictive analytics at its core and ML is simply a term used to signal that the model evolves on its own with usage, although humans of course calibrate the model as required), has been used for years. Insurance underwriters use predictive analytics/ML to inform underwriting processes. Large retailers use advanced analytics to inform site selection (modeling the intersection of weather patterns, traffic patterns and demographic data). Companies use advanced analytics to predict customer/subscriber churn. The military uses image recognition and mapping data to proactively identify nuclear missile sites. Netflix and the various streaming services use predictive analytics to power their recommendation engines. Online retailers use ML/ predictive analytics to monitor user behavior and to execute targeted ad campaigns. I can’t think of a Google product that does not leverage AI including Google Photos, Google Maps, Live Caption, Google Translate and many more. Think of the financial quant shops that use ML models to predict the next move in equities.
I could go on and on about various use cases. “AI” technology has been widely available and widely deployed for years by large companies, mid-sized and small.
It is Generative AI that is the new wrinkle. We have written numerous articles about Gen AI, most of which focus on the fact that there is much hype around Gen AI. Vendors are leveraging this hype to accelerate sales cycles and to goose the value of their equity.
Gen AI’s success or failure will be a function of end user engagement. We are largely in the build phase today. Most of the Gen AI revenue that has been created to date is a function of paying for the chips to power the LLMs, or paying for the right to license the LLMs (typically through an API arrangement). It is unclear at this juncture what level of Gen AI uptake is occurring at the end user level.
Further, the latest and greatest Large Language Model or Multimodal Model may not be required for your “AI” initiative. Perhaps Gen AI is not the answer to the outcome your organization hopes to achieve. Perhaps a cloud database with an NLP front-end will deliver the user experience you are looking for. Perhaps an off-the-shelf predictive analytics software package will deliver the customer insights you are looking to gain from your customer data. Perhaps the core AI services offered by AWS, Azure or GCP combined with a bit of your own Python or R programming may net the result you hope to achieve.
The AI hype generated by the Financial Media and vendors to propel ad revenue and valuations is just that - nothing more than hype.