There’s a great scene in the Simpsons where Homer goes to a candy convention where a salesman is hawking wax lips and claiming it's a “candy of 1000 uses.” Homer stops and challenges the guy, asking him to tell him what those 1000 uses are. The salesman indicates that you can, obviously, wear them and pretend they're your own lips. When Homer pushes for a second use, the salesman pretends to go down fake stairs because he is “needed in the basement.” (Beware, I use this trick a LOT. It doesn’t usually work).
There is a Simpsons analogy for all things in life, and this one for me represents where we are right now with Gen AI — a solution that is looking for problems. That sounds trite, even to me as I type it, but I think in legal we’ve already been through a few distinct phases with this new tech:
Phase 1 - Early 2023: The freakout.
Phase 2 - Mid-2023: Chill a bit, and learn.
Phase 3 - Late 2023: Experiment and start to build random stuff.
Phase 4 - Now: More of the same, with some firms focused on building useful stuff.
We are starting to get our arms around what LLMs and Gen AI are good at. So, what do we do with that knowledge?
My suggestion -- not a single thing. Not until it becomes useful to solve a problem. Like any other technology, you leverage it only when it's the lightest weight, most effective tool for the job. You wouldn’t bring a hammer to a pencil task, and you shouldn’t bring an LLM to provide airline customer service support via chatbot, or try to use generative AI as a project management tool, or in any context where adherence to rules is important.
Now that you know the wax lips are there, you can walk out of the fake basement with them when it’s clear that wax lips are going to be helpful in a certain circumstance. For example, if you need to create some structure out of unstructured data, an LLM may be the better tool than a parser.
Let’s get back to what we were doing before. Assessing problems by understanding them from the perspectives of all people who touch the problems, designing solutions that will best solve that problem, and THEN asking — is this technically feasible? What tech is needed here?
If some or part of this answer is Gen AI or an LLM - cool! You’re probably saving a TON of time from what you were able to do years ago (or possibly even able to do things you couldn't reasonably do years ago). But also, be okay if the solution is feasible using plain old vanilla technology. Because all that matters is that we are solving problems the best way using the correct technology. I can promise you, the fancy headline that you’re hoping you’ll get using Gen AI will be largely ignored at this point, because consumers and buyers are only interested in what problems you can solve for them and not what utilities underlie that solution.
-- Nicole Bradick, CEO
