Data has been called the “new oil” because of its value to today’s computer systems. When it comes to AI, data is critical. AI needs huge amounts of data for training.
Essentially ChatGPT is now giving us access to a portion of the knowledge out there on the Internet (which is a lot—at least 1,200 petabytes.)
For many of us our focus over the last several months has been how to develop the best prompts to get the best responses from ChatGPT. Prompts have been getting longer and more complex, some of them essentially turning into a form of computer programming language. “Experts” are designing prompts that squeeze just a little bit more out of ChatGPT each time by “talking to it” in a way that makes it work more effectively (or just harder.)
Here’s my forecast though: Prompt engineering will not be enough to give any business a lasting competitive edge.
The techniques for building prompts are already being widely shared. Anybody with a will, some time and a small company budget can learn this stuff. This information is going to spread rapidly and so nearly everyone will have it. Further, ChatGPT was designed to work with plain human English—eliminating the need to have to learn a computer language. Over time, ChatGPT and its competitors will get better at generating nuanced results without humans having to tie themselves in knots to communicate with them.
Prompt engineering will not be a key competitive advantage. Data will.
Bloomberg just showed us the path with its new AI tool. This tool is based on Bloomberg’s huge store of financial data. This is the beginning. We are going to see companies figuring out how to better use and monetize their proprietary data using AI.
In sales a lot of our proprietary data lives in our CRM system(s). We’ve just seen the first waves of generative AI tools for CRMs come out from Hubspot and Salesforce.com. These are very early versions and we will see many more interesting applications here.
I am looking forward to the day when I can ask a ChatGPT type interface “Who should I call next and why?” The AI will help me figure out:
- What trigger events have occurred to indicate it’s a good idea to call this person now. This can come from internal “signals” like an email from a client, or a prospect visiting my website, or external signals like a news story.
- What I can say when I call, based on any information that is available internally or on the Internet.
- How can I best connect to this person? Find the best “relationship path” to them—based on Linkedin data (not very reliable) and all internal data, like all company emails and CRM records.
This AI help will only be useful when the data in my internal systems, like my CRM, is accurate; otherwise, the AI will give me nonsense results–“garbage in, garbage out”. I suspect therefore one of the key applications of of AI will be data cleaning.