smartphone ai apps

LLMs and Intent Laundering

I spent nearly five years working in the financial services industry. One of the recurring themes for anyone in corporate work, and especially in the financial sectors, is computer based training. We called them CBTs, because who doesn’t love an acronym?

With some back-of-the-napkin math, I think I learned about Money Laundering at least 20 times. Now, not how to clean my money, more about how the bank had practices in place to prevent it. I think that the (admittedly inaccurate) visual of a washing machine with a bunch of money tumbling inside has been a great way to help me understand just what about AI content is usually so off-putting.

row of laundry machine
https://unsplash.com/@von_muenchen

In an action movie, money laundering looks like the bad guys busily cramming illicitly gained money into a row of washing machines. The idea is that this illicit (new) money can look like it’s been circulating and is thus less suspicious when it’s later spent.

Whenever anyone sets out to communicate, they begin with intent.

Intent can be simple. My kids get home from school and ask, “Dad, can we play Fortnite?”

Intent can also be obscured. Ever found a really helpful comparison of products designed to solve a problem you’re facing? What happens when you get to that part about 1/3 of the way through and realize the company blog is now recommending their product?

Intent is always the beginning, and for years the way you work that intent out has been where the magic happens. Some people are excellent at selling, at getting the message across, at balancing perfectly between intent and relevance.

Large Language Models (AI, LLMs, etc) have proliferated in the past few years. Lots of words have been written about them and how they work. More words have been generated by those tools.

smartphone ai apps
https://unsplash.com/@salvadorr

LLMs take their huge data sets and when creating content for you, they use their computing power to give you something unique that is informed by all of the patterns they have been trained on.

When you want something unique that does not have to be original, this is a huge positive! If I wanted to research “10 cars that have three rows of seating”, I would be happier to get a list of 10 popular cars, not 10 oddballs.

Consider however, what I’d get if I wanted to create ideas for social media posts for a roofing company. LLMs, by design, are going to spit out something that’s the approximate average of what it was trained on. Roofing is fairly localized, so maybe it’s okay that a roofer in Dallas has Facebook posts that sound a lot like a roofer in Los Angeles.

It’s easy to look at these things and say, “eh, good enough.” But, the content that gets created by an LLM is often just that: good enough. It’s not great.

I’ve seen some really helpful use cases for LLMs, but when I’ve sat down to edit long-form content it’s been a frustrating experience. I can’t quite put my finger on it, but it’s just different. In a bad way.

The reason is hard to pin down, but I think it comes down to intent:

When an LLM generates anything it tumbles your intent around with everything it’s been trained on, laundering your intent as it creates.

Intent Laundering

The part that gets missed is the part that has perplexed every single person who has said something and been misunderstood. Your intent when you told that joke was to make someone laugh, but instead they looked hurt. You meant to make the sale and when you pushed the shopper to make a decision, they hung up.

When your intent goes into an LLM, picture a $1 bill going in with thousands of other bills. The bill that comes out is going to look like the thousands of others it has been tumbled around with. It’ll smell like it, too.

When your intent to make a Facebook post for your roofing business goes in, it gets tumbled around with all the other Facebook content the LLM has been trained on. It comes out sounding, smelling, and looking a lot like the other content like it.

What’s getting lost? Your intent. It’s often hard to see because we, as humans, don’t always think through exactly what it is we intend before we do things. My wife and I love quoting movies and shows to each other, and I’ve had the words, “That’s what she said!” (Michael Scott, The Office) pop out at a few rather inappropriate times.

When content gets off track is when the thing that you meant, your intent, gets tumbled around with all of the things other people meant when they put their words out there. Namely, those words are the data the LLM was trained on.

An LLM doesn’t know anything about intent. It knows that this word follows that the majority of the time, so it strings them together and, often, it makes sense.

How to Mitigate Intent Laundering

So, where’s my dollar?

As a writer and content strategist, I have a few tips for pushing back against this loss of focus in a piece of content. No matter where you are in the process, whether you’re looking at a rather bland blog post, or a snazzy slideshow, this is what you do:

Stop, minimize the window or turn off your phone’s screen

First things first, you need to turn away from the screen! If you can, silence notifications and reduce the “urgent but not important” interruptions.

Take a moment to breathe

I’m telling you, giving yourself a minute and just clearing your mind is huge. If your thoughts are bubbling over, write them down so they stop intruding.

Ask yourself, “What am I trying to communicate?”

If it’s an email, a blog, or a Facebook post you have a reason for putting it together. You’ll need to get crisp on this.

It should be one intent, one message

We’ve all been there. Content is most effective when it tries to do one thing. Is it informative? Let it inform. If you want to convince people to buy, end with a single clear call-to-action.

Ask, “As a whole, what intent does this content communicate?”

Now, with a little mental distance, look again at the screen and ask yourself what that content is saying.

If there’s a mismatch, start your revisions by pointing back to your intent

Sometimes you may want to start over, but often you can begin revising a piece of content once you have a clear idea of what you intended in the first place.

(note, I’m always a fan of editing things myself, but even going a few rounds with your LLM of choice is likely going to help.)

several used dollar bills laid out on a table
https://unsplash.com/@emiliotakas

Hang on to Your Dollar

Friends, the era of LLMs has been thrust upon us and it’s been a bumpy road for a lot of people. For many, these tools have allowed people to build and create in new ways. For some, this has meant the loss of jobs and opportunities. Don’t even get me started about the amount of money flying around in this sector.

What I want to encourage you with today is this: hang on to your dollar. Take the time to think about what you intend to say, and make sure that your content says that.

Don’t let the laundry machine eat your intent and give you one that looks a whole lot like everyone else’s.