Finding More Grit: How Generative AI Will Transform Talent

Speaker 1 (00:00):

Please welcome to the stage Allie K. Miller, AI entrepreneur, advisor and investor.

Allie K. Miller (00:17):

I think I ate so many beignets yesterday. I asked Michael if he could have a wheelbarrow backstage just in case, but I made it. So, very excited to talk about AI. And I'm here to say that you are in the dawn of the AI age. And just like Angela said about the importance of growth mindset, I need you all to apply that growth mindset to technological innovation as well. So I want to walk you through how you can take advantage of this moment. Because if I could sprinkle on a little FOMO, you've got less than a year to figure it out.

(00:59):

So every big speech, every wedding speech starts with an ancient Greek quote. And so this is from Heraclitus who said that the only constant in life is change. One thing that I think that this quote gets wrong or is missing is that change itself is changing. The rate of change is ever-increasing. And so though you've seen change in your work, you've seen change in your personal life, you've probably changed jobs, change cities, this feels different. There's a shift happening in the world of tech and AI. And I don't know if you guys feel it, but it feels a little like acid reflux to me, though it might be the beignets. We don't really know.

(01:47):

And so I want to be able to really share this one most important thing, which is not a single AI use case, not a single AI technology. It's to understand how fast this space is changing and how it is bringing about a massive business shift that you guys will be a massive role in that change. So I'm going to break down in four different ways. The first is scale and what is happening in tech. This is the first of four. We're seeing a bunch of different apps and products there and how long it took, how many months it took for those apps and products to hit a hundred million users. So it took the internet seven years. It took WhatsApp, which came out 40 years later, half that time, three and a half years.

(02:38):

And can you guess the top two? I'm going to guess that you guys can guess the number one.

Audience (02:38):

ChatGPT.

Allie K. Miller (02:44):

ChatGPT. Yeah, we all heard the choir. Got it. So ChatGPT took two months to hit 100 million users. And a lot of the news articles heralded this as this unbelievable exception to the rule. But what a lot of people missed from that news story was the second-fastest app to hit 100 million users. Any guesses? Google. I heard Google. Second fastest was a different AI product called character.ai. Anyone in this room tested it out of pure curiosity? I'm going to guess your kids have. It is one of the fastest growing AI apps among Gen Z and Gen Alpha.

(03:26):

It also is happening in terms of performance. This rate of change is ever-increasing. This horizontal line is human performance on certain tasks. And as that line crosses that horizontal line, that means that AI has beat out humans on that task. So for handwriting recognition, which came out quite early, took 16 years to outperform humans. For image recognition, which came out later, it took six years. For reading comprehension, one year. Those slopes are ever-increasing. I think you're picking up on this pattern that I'm calling out.

(04:02):

And the biggest thing to call out here is that it's not just about performance, it's also about scale. And so I'm very good at image recognition. If I see a photo, I know it's a zebra. If I see a photo, I know it's Michael Bush smiling or calling out growth mindset. The biggest change is that I can label one image a second and AI can label 10,000. So you also have to factor in that scale. And we see it also on common benchmarks of multi-task performance. This is a common benchmark called MMLU. And we went from subhuman to superhuman performance in four years.

(04:45):

Now folks in this room, I think a lot of you guys probably have five-year roadmaps. And imagine that this change happened in one span of five years. So again, the rate of change is increasing, but that doesn't matter if it's only the big companies that can take on this change. Well, despite these models getting bigger, they're also getting more efficient. You see models that cost thousands of dollars years ago now can be done for less than $5. You've got students building massive ML models that only research facilities could take on a couple years ago.

(05:24):

And the last is accessibility. Michael talks a lot about building a future for all. And so again, AI innovation does not matter if it's only the precious few. And it started with the million of the most technical learners. It was machine learning engineers and data scientists. Those were the only people that could access this technology. It grew to 28 million when developers had access to it. I remember testing out GPT-2. This was in 2019. You could write the beginning of a sentence and it would complete the sentence. And I just knew that the world had changed. And that was five years ago. And now people are making movies with AI, right?

(06:08):

So this change is now hitting all internet users. We are now at 5 billion people that have access to this same technology. The big question is how are you're going to take that vision and turn it into action. Because I could get up here, I could show you 7,000 graphs. It doesn't matter unless you do something about it back at your company, back at your team, back at your executive table. We are on the precipice of one of the most significant changes in the history of business. And I know that sounds hypey, and I know I sound like a crypto shiller, but I assure you it's true.

(06:52):

So it's turning that vision into action. And I want to share with you the big thing I learned at Amazon. So my role was the global head of machine learning for startups and venture capital. I worked there for almost four years. I worked with every top AI startup, AI founder, AI researcher. I lived with the founders of Anthropic, I worked with the founders of Stability, et cetera. The big thing that I learned in talking to these founders was that on the surface, in TechCrunch articles, it looked like they were just building AI products.The big change that I was seeing was what was happening inside the company, the things that more are going to affect the people in this room because how they were running their business had completely changed. This is a global response shift.

(07:45):

And so it wasn't handling their business with a traditional model because that's inefficient scaling. So we're going to cross that one out. It wasn't handling the rate of change with a digital native business like Netflix, Airbnb, Pinterest. That made sense as a business model at the time, but now doesn't give you that same level of adaptability. So we're going to cross that one off too, right?

(08:13):

The new business model that is happening inside of these companies that are now impacting enterprises and the folks in this room is that they are building an AI-first business. And so if Angela's word is going to be grit, I'm going to say the word AI-first about 7,000 times in the rest of this talk. The breakdown, I first want to level set on AI. A lot of folks know about generative AI. How many people have used ChatGPT out of curiosity? We got a straight-hand, pink blazer. Love it. Love it. And how many people have illegally used AI tools that their IT department didn't approve of? Straight-hand right here, black shirt, I love it. I love it. I'll hire you tomorrow.

(08:59):

So generative AI is the ability for an AI system to create something net new. It might be text, it might be audio, it might be a song. You can now create AI songs on tools like Suno or Udio. It might be an entire movie. It might be a laugh track, really doesn't matter. It is about creating something net new. And generative AI itself is not new. It's been around for decades. But bad generative AI has been around for decades. Low quality generative AI has been around for decades. The reason that now is different as you saw in those graphs is that we're at that hockey stick moment, which kind of actually feels like a vertical line at times, but mostly a hockey stick moment.

(09:39):

And some examples that pink blazer and black shirt have clearly seen, first is the ability to create images, videos, even worlds. This is an example from Midjourney, but you can create your own AI images in Canva with Adobe or Kore.ai.

(09:58):

Anyone heard of world models out of curiosity? a lot of the research companies right now like NVIDIA and Meta, they're trying to build 3D representations of our entire world to mimic the physics of every single atom around us. So I'm just going to flag that now.

(10:15):

The next is about decision making. And I'll show you an example that my team has actually built that we use to make business decisions. Every single large decision that we make in a brainstorming setting, we then have an AI system come in and answer four questions for us. "Are there any financial considerations we haven't considered? Legal considerations, user considerations and product considerations?" And so at the end of every single call, and we record all of it, we send in that transcript, we ask those four questions. So we're building a more robust decision-making process.

(10:49):

We see it for copywriting and synthesis. I'm sure many of you have written AI generated tweets or AI generated LinkedIn comments. But synthesis is also a huge use case. I was using Google Gemini 1.5 Pro, I fed in every single Amazon shareholder report for the last two years and I asked it the question, "What was in 2022 that was mysteriously missing from 2023?" And in less than 40 seconds I got the answer. Now it might be lying. We don't know. AI has an unbelievable skill set in terms of hallucination. I asked an AI system about myself, it told me that I had three children. I'm not even familiar with one. So be sure to validate those answers.

(11:42):

Next is code and agents. A lot of people are using AI models to draft initial code pieces. And the last is text to everything, text to image, text to video, text to song, but that is also going to open up anything to anything. The ability to upload an image and get code. There's already been demos of people sketching out a website on a napkin and immediately sending it to an AI system and that website is coded and ready to use.

(12:11):

So my prediction here, by the way, is that I think we're going to be able to somewhat meaningfully talk to our pets in the next 10 years. So-

Audience (12:12):

Whoo!

Allie K. Miller (12:21):

There was a "whoo!' You're wondering what your cat is thinking about you at night? Okay. So those aren't exactly use cases, right? Those are just tech uses. And so a very quick explanation of traditional AI that many of you guys have probably used in the last decade. Things like analytics, things like image classification or things like forecasting. Maybe you're forecasting for your head count for next year. That is different from generative AI, which again is this net new creation. It might be a synthetic data set, it might be a new image, it might be an entire business plan, but this is that different. AI is not new and both sides of this are unbelievably valuable to your business.

(13:08):

In terms of HR use cases, this was shared at the beginning of last year in terms of what use cases are getting the most traction in HR. I wanted to be able to share this with you all. It kind of starts with employee records management. These are all kind of going through to recruiting and hiring, cuts through to performance management. This was an interview of 250 of the top CHROs.

(13:32):

We also see, and I'm going to kind of share with you guys a secret, I told Michael I was nervous to share this but we're going to say it anyways. HR is often the detractor in the room for AI. There was a light murmur. I heard it, I felt it. It's an important detractor. So I think that just like Angela said, which is having both of these things working in action, right? System and individual. It also is required that in your exec team there are pushers and pullers or generators and discriminators that are setting that path. And so HR is kind of in the middle of the pack in terms of roles and how fast they're adopting AI. You guys are a lot faster than legal, don't worry. But marketing's got you beat. One marketer snuck in, guys. Security. Security.

(14:30):

And in terms of the most used software, this is really breaking it down by task in HR. The top is more around writing and data analysis. The middle of the pack use cases are a bit more around admin or resume reviews or job description writing. And the bottom the most complex and probably the biggest lag is things like back office or payroll or benefits. I'll add here. I see a lot of phones coming out. Take as many photos as you want. We're going to be going through the rest of these slides fast.

(15:01):

All right, so how are you going to manage this? We're already seeing that the shift is happening. Allie's got acid reflux. How are you guys going to manage this and be a very vocal presence in these rooms? What is happening is that these folks are building an AI first business and I want to share with you the three Ps of what that even looks like. I have had hundreds of conversations with some of the top C-suite executives in the world about how this looks in practice. And so I want to distill that down to a framework for you guys so you can take it back to your businesses.

(15:38):

The first is people. This should be screaming out your name. The second is process. And the third is product. AI is infused in all of these. Again, it's not just adding AI, a sprinkling AI into your website, adding a chatbot. What's happening underneath the water, the 90% of the iceberg, that is the big piece. So when it comes to people, this is about supercharging your employees. What does that often look like? It's about employee productivity. And it's not just about saving time, it's about morale boosting and employee engagement boosting. Microsoft yesterday just came out with new statistics that AI power users feel that they're more creative and more motivated at work and that they can better handle an overloaded workload.

(16:33):

So I want to share with you one big example for each of these categories. And this example is going to be from Morgan Stanley. What I love about this example before I get into it is that it is still internal only, right? They're derisking it away from that user. And I spoke with the head of AI at the New York Times and he had a wonderful quote, which is that it starts and ends with humans. So this example shows that.

(16:58):

So Morgan Stanley decides to give this super souped up chatbot to all of their wealth managers. So if I'm a wealth manager, I'm having a conversation with my customer, they ask me a question that maybe I don't get all that often. Maybe it's something like how do I set up an account and I get to go to my super secret chatbot, I get to ask a question, I get the help, and then I look really smart with my customer, which is great because they got the help that they needed. They are happier and they didn't have to learn a brand new AI system in the process. They had a human to human interaction. The trust is still there, the culture is maintained at that company. And again, Morgan Stanley only had to train their internal employees. They started small, they started with the wealth management team and now they're scaling it out to the rest of their company.

(17:47):

AI is a leveler for people who are maybe underperforming or maybe more junior in your companies. Okay? So BCG has run this study and they found that consultants using AI finished 12% more tasks, completed them 25% more quickly and produced... You guys are like, "These are probably generic tweets with the word delve 100 times." It produced 40% higher quality results. And so this is going to have an impact on how employees are getting measured and how employees are actually able to perform.

(18:30):

We also see AI impacting process. This is about supercharging your operations. This tends to be back office, data insights, streamlining your ops decision-making and interoperability. Again, the first one, people. And this one, process, are a bit more on the cost saving side. Less waste, more efficiency.

(18:52):

So this example has nothing to do with HR, but is a very important takeaway for the folks in this room for how you are thinking about what types of stress or changes your managers are feeling. So this is a company called Pactum AI. They're an AI negotiation bot. And so white is Pactum. White is an AI bot. Blue is a vendor. So they're able to have this contract negotiation back and forth. So Walmart decides to run a POC with this. And I saw Walmart on some of the slides, so maybe there are Walmart folks in here. But Walmart decides to run a three-month POC on this. This is sort of before the recent ChatGPT craze. So they send out a call to action to their long tail of vendors. Not the huge, huge vendors that they have these massive deals with. Those are still people relations. They send it out to the long tail of vendors, things like shopping cart procurement or janitorial services. And they send out 100 calls and they say, "Who wants in?"

(19:59):

How many people do you think said yes? Kind of creepy. It's an AI bot.

Audience (20:03):

10.

Allie K. Miller (20:03):

10?

Audience (20:04):

100.

Allie K. Miller (20:08):

100? Oh, my god. Big data person. The number of people that accepted, 89% of these vendors opted in and said, "Yes, I wanted to negotiate with an AI bot. Let me play." They were hoping to have a target close of 20%. And again, Walmart sets up the guardrails for this. They say, "I'm willing to sign between this contract size and this contract size. I'm willing to have these net pay terms." So they are still in control. It's just the process part that has helped with AI. So they had a target close, 20%. Target close final, 68.

Audience (20:43):

Wow.

Allie K. Miller (20:45):

Wow indeed. I'm going to show you two more stats that are going to make you say wow louder. Here we go. It does not matter unless they saved something in the process, unless they did some sort of cost savings. And so this process ended up saving 3%, which if you're Walmart is a heck of a lot of money. And they went to these vendors that they were negotiating with and they said, "Hey, what did you think of that process? Did you like it? Do you want to do it again? Did it feel icky?" And they went to get this feedback and they were shocked at what they found. Because when they asked customers, "Did you prefer negotiating with the AI?" The number of customers that said yes was 75%.

Audience (21:33):

Wow.

Allie K. Miller (21:34):

That was the louder wow. Okay. The reason that this is so important is again, it was not automating the entire negotiation process. They still maintained the deep customer relationships. It was the long tail of vendors that had been forgotten. It was the people who their emails weren't getting responded to and they had to wait two weeks just to hear back. This was actually a trust building exercise for Walmart and they got to reallocate their people to the most pressing problems.

(22:05):

When it comes to people and business strategy, this is a real conversation that I had with ChatGPT. So I gave it the prompt. You're an expert in HR and recruiting. I described a little bit of my business, said I advise Fortune 500s, I invest in startups, et cetera, et cetera. And then I said, "I'm hiring several people. Can you kind of break down what their roles and responsibilities should be? Can you tell me the key differences between these roles. And I want to hire someone else. Can you just figure out what that next person should be? What is the role that I should have? What should be their responsibilities?" So I'm giving it this kind of blank head count and it comes back, gives me a rundown, and it says, "Here's what a COO does. Here's what a content manager does. Here's what a PA does. And that last person should be some sort of admin support or ops support."

(22:56):

Now this answer is quite generic, but it made it so that I wasn't starting with a blank page. I tell everyone I am allergic to blank pages now. I cannot write a business memo from scratch, write an email from scratch. It now feels like etching into a tablet. I mean it feels very antiquated.

(23:22):

So it gives me this rundown and then I ask it follow up questions and I say, "Based on my business goals, how would you then shape my business strategy and people's strategy?" And again, this is not going to answer 100% of my questions, it's just giving me something to react to. It's a brainstorm buddy. It's a co-pilot buddy. And so it gives me that rundown talking about "Maybe you should hire a product manager. Maybe you should hire a designer for upcoming courses." So people are using this as a copilot for any sort of task and this is going to have business management impact.

(23:58):

And the last is product. AI is supercharging solutions. The flag here is that this category, the product category is the highest risk category because it's external facing. Everyone in this room has an unbelievable amount of trust with your customers. Do not, in the hope of something cool with AI, risk that. As much as I get up here and talk about how cool AI is, this is not me saying to use AI in every single thing you do in your business.

(24:34):

So what does it mean to supercharge your solutions? This is about growing top line revenue. This is about factoring in AI into your innovation process, into your product development process and to deal with ever-changing market demands. This is about adaptability, improving quality. So an example from Stitch Fix is that they use AI to mass generate product description drafts. They have hundreds of examples of reviews. So let's say that they have a dress that maybe they thought was really good for weddings, but all the reviews are like, "Nah, this isn't really a wedding dress. Better for work." They're able to take all of that plus a model that they fine tuned to be in a Stitch fix voice. And they then said, "Generate tens of thousands of product descriptions." And so the time that it took to do this, it went from two weeks down to a minute. And I'm going to call out here, AI is not perfect. It was not 100% approval. It was a 77% pass rate. So these expert copywriters, instead of writing 10,000 from scratch, they became an editor. They became a manager of the process.

(25:48):

And so specifically for the people in this room, I want you to feel that internal shift that's going to be happening for different roles, whether you're a copywriter or an engineer, is that what it means is that humans are going to be taking on more complex tasks and might have actual implications on things like compensation or where you might hire from. So Stitch Fix does this. They have the experts check it.

(26:15):

Another example from Stitch Fix is that they use AI to generate new outfit combinations. This was picked up by me, I just want to call that up, but I'm sure AI could have told me something better. They generate 13 million new outfits every single day and 43 million new outfit combinations every day. And I think a lot of people in this room are probably like, "Who cares? I'm not in retail. Onto the next example." But the call-out here is that this expand beyond this use case. The call-out here is that the future of our businesses powered by AI is hyper-personalized. Every single thing is going to be a market of one. Every blog post you read will be a version for you. Every ad you see will be a version for you. Every email you get will be a version for you. And you cannot do that if humans are writing every single thing from scratch. That hyper-personalization which builds trust and builds conversion, that is going to be powered by AI because that's the only way to hit that scale.

(27:22):

One last example is from Audi. They're using AI to help design wheels. And so what they actually did is they built up two different models. One model is sort of the very creative outside the box Met Gala sort of thinker who is throwing out crazy ideas for tires. It's a square, it has spikes coming out, they're pink, whatever. And the second model is the discriminator. It's looking at that going, "That one's not Audi. That one's not Audi. That's got good Audi vibes. Let's bring that in." And so they're able to balance two AI models to build this whole workflow. And so they have designers looking at what that output is, the ones that this model kind of pre-approves of. And then the human is in control of this process and gets to go back and forth and says, "I know Audi best. Here's an edit I need. Here's an edit I need,. Here's an edit I need." And so instead of 20 revisions, they're able to take it down to say four. And then they auto-prototype it using 3D printing. And now they have an end-to-end workflow with AI.

(28:33):

The big takeaway here is that everyone that I see at these events, and I've talked to some of you in the elevator sneakily, is that people believe that it's just AI in a product and the call-out is that it is happening in every part of these businesses and is often not these public use cases that are shared. So there are many use cases across these three, whether it's people or process or product. It could be everything from managing invoices, it could be using platforms like Mentra to find neurodivergent candidates. I heard Michael and Angela talking about neurodiversity. There are now net new AI platforms that are helping recruiters do much more niche candidate sourcing.

(29:22):

There are platforms like Fountain that are helping people build better employee engagement and to look at past exit interviews to build better processes. Culture Amp reduced churn at NASDAQ by 50% using an AI platform called Culture Amp. So this is what's happening in this space.

(29:44):

I do just want to share some final thoughts, and good news is that Angela and I have very similar thoughts on how you guys should move forward in this. First, which is the AI-specific component, is that whether it comes to people, process, or product, you're going to feel three things as HR and people leaders. AI is a leveler. We saw that BCG example, that some of your underperforming or less skilled employees are going to be able to be ramped up much more quickly and that's going to have performance management impact. We saw that the way that I use ChatGPT, the way that Walmart was doing negotiation is that people are using AI as a partner, as a co-pilot, and that's going to have business management impact. And we saw AI as a transformer from Stitch Fix and Audi that it's going to have talent and revenue impact because the task that people are taking on will change.

(30:37):

For the people in this room, start small. A lot of this can feel overwhelming, right? Morgan Stanley only started with one team. A good rule of thumb is just start with 2%. Start with one little tiger team of couple people like a task force. AI is a tool. It is not a magic silver bullet. It's not going to solve your problems. Always start with the problem to be solved. Do not just AI-ify things and think that it fixes your business. Remember that data is the source of everything. And as you're thinking about implications down the line, trust and transparency is key.

(31:15):

The last is that I agree with Angela. It all starts with culture and trust. And you guys are the heralders, the keepers of culture in companies. And so unless you are bought in, it doesn't happen. If you're wondering how I've built an AI-first team and you're wondering, "Hey, that prompt looked good, or the way that she talked to that AI system was interesting," this is a totally free video that my team put out where we are literally showing how I run my AI-first team. This is not me saying you have to do it this way, but it is going to give you those tactical pieces that you can take home. So again, you're in the dawn of the AI age, the rate of change is increasing. I ate too many beignets. These are the top takeaways. And it all starts with culture. Thank you very much.