Advanced Generative AI Products Business Leaders Must Pay Attention To Today: Intelligent Advisors
Part 1 covered the first wave of Generative AI products, agents. Agents will improve their functionality over time to become what Bill Gates sees as disrupting companies like Google and Amazon. Intelligent agents will bite into Google’s search business, but not enough to put it out of business.
Three improvements will drive the increase in functionality and utility. First, businesses will get better at selecting data sets and use cases for LLM finetuning or retaining. Intelligent agents are high-value prototypes that will reveal the paradigm to a broader business audience. The more business leaders “get it,” product and data teams will build. There will be a learning curve, but with practice comes proficiency.
I’m not as hands-on as I used to be, so I’ve gotten the chance to watch how quickly data teams improve after 2-3 implementations. It’s difficult to see the change when you’re part of the team, but repetitions create significant proficiency in short time spans. The ChatGPT moment has created an unexpected consequence. Data teams are quickly becoming product and delivery focused.
Generative models will become multimodal around this time next year. Adding images and audio to prompts and responses will enable intelligent agents to support more advanced types of human-machine interaction. Google and OpenAI plan to deliver multimodal products in about a year.
People increasing their usage of intelligent agents will curate data sets that enable hyper-personalization. There are privacy issues to overcome, and many users will not adopt an agent until those are resolved. Finetuning must improve so the new or updated weights can live on the user’s device.
A few solutions are already being worked on that have yet to be released. This is one of the early independent projects. I don’t expect this challenge to take more than a year to overcome, so a viable solution for mobile devices won’t hold up progress.
Those advances set the stage for agents to transition into more capable utilities. Advisors improve upon agents by offering predictive, prescriptive, and diagnostic guidance for a small domain. Agents do what we tell them to, while advisors help us decide what to do ourselves and help us execute.
There is some speculation that advisors will rapidly generalize across domains. I think that’s more than 2 years off. The applications I see for intelligent advisors extend the search paradigm and serve many underserved use cases. Generalization will happen eventually, but I won’t cover that type of intelligent assistant in this article.
Searching For Expert Advice
Some searches are simple and well-served by current search engines like Google. Complex searches include those where the language needed to explain what the user is looking for doesn’t fit the search box formatting. Another category is a search where the user doesn’t really know how to express what they are looking for.