Connecting a Few Dots in AI for Marketing

A few thoughts on AI and marketing from recent events and conversations:

1. ‘Prompt engineering’ is a thing – and you should know what it is

At the PepperContent session in SF I learned that prompt engineering is about to become a job description and there are marketplaces popping up to use, buy and sell prompts – beyond your LinkedIn feed... Checkout promptbase.com for Dall-E, Stable Diffusion, MidJourney and, of course, ChatGPT prompts. If you don’t know these names take time to learn the story behind them because there is a fundamental tug of war in approaches to Gen AI (think iOS and Android) – not surprisingly it’s coming down to open vs. closed approaches.?

1. Open vs. closed approaches will matter more and more

Facebook and Stability AI are staunchly in the open camp, with Meta opening things up with Llama and Llama 2 and Stability AI creating revenue streams by applying their models to enterprise-client-owned data sets. ChatGPT is closed and their dataset is a big black box. Authors and writers recently sued the parent company OpenAI in a class action suit after those same authors found evidence of their works showing up in ChatGPT results.

Open systems mean potentially significant competitive advantage for those enterprises that figure out how to apply a large language model to their proprietary and highly valuable datasets.? There is also very significant potential for abuse of these technologies. This type of control over the training data means you have increased confidence the results are accurate because you control the quality of data going in to train the model. And the dataset doesn’t have to be enormous to be valuable.

Maybe LLM should stand for larg(ish) language model..

2a. Know your data sources

The closed ecosystems are fun to play with, but buyer beware. OpenAI is based on a dataset from Dec 2021. And some AI tools just flat out say their tools may be wrong.?

Here’s an example:

When my team was using an AI tool to ideate names for a recent branding project, we discovered some great ideas, and the tool definitely accelerated our ideation phase. However, when it came to making decisions based on the output, I saw a small little interstitial message that said “Warning: AI generated feedback may not be accurate.”

What??

So I tested it on one promising name based on a Nigerian word for “connection.” Turns out after consulting native speakers, language dictionaries based on all the prominent languages spoken in Nigeria this word never showed up in any search related to connection. Buyer beware.

1. Legal and marketing need to be well aligned

One comment I heard repeated time and again was how the lawyers were going to squash GenAI before it got out of the gates, due in part b/c of situations like the above and worse – if engineers upload source code to a cloud-based GenAI tool to accelerate code development the company effectively loses control of that proprietary codebase. Spend some time partnering with your general counsel and help them understand and navigate the choppy waters of GenAI – you’ll likely help preserve a license to operate.