I have yet to find a company that has successfully scaled genAI in their organization.

No shade - it’s super hard.

The reason it’s hard? We are using an outdated paradigm.

Consider this:

A typical digital transformation is like corralling your workforce onto a Carnival Cruise. The initial logistics are a nightmare, sure, but at last everyone’s left the port and is now sailing off, together, to Virgin Gorda or wherever.

Generative AI, on the other hand, is like giving everyone keys to their own jet ski. With zero instructions. And zero destination. And some people are still back in their hotel rooms.

That’s because generative AI is wildly individualistic. A digital transformation improves an old system. GenAI, though? That improves an old YOU.

How do you scale something that works differently for everyone?

You need a framework. The one I use with senior leadership looks like this:

TRAIN - CHAIN - CODIFY - SCALE

Let’s start with the problem we’re tackling.

There’s no one single approach to using genAI, because the data set your people will be using will be…wait for it… their own brain.

Which means outputs will be infinite. (Actually, the number of outputs is equal to the number of people in your organization. But ‘infinite’ sounded so cool.)

Making this even more difficult?

It’s extremely hard to teach. Even though everyone knows it’s powerful, even though it is theoretically simple…it’s just hard.

So it’s less like ‘Here I’ll show you how to use this new sales software’ and more like ‘You should really do yoga every morning.’

Anyone ever try to convince you to do yoga every morning? I have. And I know it’s healthy and I’ll feel great and probably live to a hundred. And I will never do yoga every morning.

That’s most people I meet with ChatGPT.

I’m like “It will change your life!” and they’re like “Okay I’ll try it" but then they don’t, any more than I’m going to wake up at 5:30 to do camel pose.

That’s why our first step in scaling has to begin with Train.

STEP 1: train

I beg you - stop relying on the ChatGPT super users in your company to train others.

The super users are are rarely good teachers. If you’re learning basketball, you don’t want Michael Jordan. You want the 8th grade basketball coach. And your favorite college professors weren’t the biggest brains, they were the best at helping you understand new information.

You need real, actual training. (I do a lot of this.)

In the jet ski analogy, it gets everyone together, gets everyone turning the keys together, and makes sure everyone knows how to ride them with equal proficiency.

A warning:

Teaching does not mean showing people what genAI does. That’s like showing somebody which way to run on a treadmill. They know, man.

And don’t teach through use cases. That’s like teaching French by using a phrase book.

You need to inspire them. Get creative. Give them a framework to think differently (this is what I do with the AI Mindset framework).

Invest in training first. Otherwise everything else will be a waste of money.

STEP 2: CHAIN

Proficiency with genAI is step one. Using it correctly and consistently is step two.

One of the biggest problems I see with folks using generative AI is that they get a great answer, then move away from ChatGPT to ‘go do the rest of their work.’

That’s because we have what I call Google brain. We are locked into this 'command - response - walk away' mindset.

But to scale your work, as an individual, you need to start chaining together tasks. You need to create an AI-powered workday.

That means completing one task, and immediately using ChatGPT to do the next task.

If you’re in HR, for example, use ChatGPT to write a job description. But don’t stop. What’s the next task? Interview questions? Onboarding manual?

Get genAI to help with every task, one after another.

Now you’re cookin’.

STEP 3: CODIFY

Steps 1 and 2 dealt with the individual.

But to scale, you need a system that everyone is using.

Forget prompt libraries and use cases. They will be useful eventually (especially across tight teams with similar tasks). But they are most definitely not good when you are first learning.

The Codify step means scaling your team up together. This is your little flotilla of jet skis with a group of friends.

Every team can come up with their own process for using genAI and - importantly - holding each other accountable for using it. That’s one of the hardest parts because it involves a change in behavior.

Team meetings should include an open browser with your LLM of choice operating as another teammate.

Your team should come up with a rules, guardrails, expectations for use.

Put them down on paper. The whole team should know and reference them.

STEP 4: Scale

Now you’re ready.

All those flotillas of jet skis are going to come together. It’s gonna be amazing.

This is where your AI Task Force comes in. I do this with companies- building out AI Task Forces. That’s where you centralize. It’s where you hear from all the teams.

What is everyone doing? What are the best practices?

What behaviors or guardrails should be made into company-wide policy?

This is important:

You don’t have to adopt each other’s models across teams.

Every team will have its own culture, it’s own way of doing things, like everything else.

But.

For this to go right, there has to be a benchmark, and it has to be set at the top of the organization:

What does a new 8 hour day work look like? What does the work week look like? What can you achieve together?

When you see all the teams working together, it’s inspiring.

That’s it!

Train - Chain - Codify - Scale

We got this, people.

AI NEWS OF THE WEEK

1. Hume gets all emotional

Hume AI is making waves with it’s Empathic Voice Interface (EVI), claimed to be the world's first conversational AI with emotional intelligence - it can apparently detect tone, emphasis, and pitch in the user and respond accordingly. This could get interesting.

2. OpenAI and Microsoft building a $100b super computer

Named Stargate (of course), the tech giants are planning on building a massive AI supercomputer data center to support the creation of future advanced AI models. There seems to be no ceiling here.

3. I, Conor

This is cheating, but I wanted to share my own news - I’ve stepped down as Dean of Students at NYU Stern, but I’m staying to build a generative AI program! Just thought that was kinda fun.

Generative AI Tips

If you’re still using Google Search - well, I would just encourage a pause.

Tools like Perplexity AI and You.com are absolutely fantastic. Not only do they search multiple sources at once, and give you an answer in natural language, and cite sources, but they also have LLMs like ChatGPT running them, which means you can put the output in any form you want.

My personal favorite is Tell me what’s happening with this OpenAI supercomputer thing and write me a LinkedIn post about it - that kinda thing.

Do I copy and paste it? No, because I want my posts in my voice.

But boy oh boy if it doesn’t save me a ton of time.

Try it and let me know what you think.

That's all for now, friends! See you next time.