Introduction

I'm Bryce. I originally come from France.

From France to Silicon Valley

I worked in mass consumer goods, I worked in consulting, I had my own company for 13 years, and then Google snatched me from France and brought me to California in 2016, where for five years I was the global head of transformation programs for Google.

And this is where I met the leaders of Moderna during an executive briefing session, it's called like this, and they blew my mind.

They blew my mind.

This was way before COVID.

A Serendipitous Meeting with Moderna

And when COVID happened and I heard that they had a vaccine for it, I kept their numbers and their emails. So I sent an email, I said, do you need anything? And they said, yeah, we need a lot of help. We need it yesterday.

And two hours later, my children were registered at the International School of Boston. And a week and a half later, I was here in an empty plane because beginning of Delta.

A New Chapter Begins

and so crazy life adventure in 2021 and this is a very special day and week for me because I'm I rendered my resignation tomorrow now on Tuesday my last day will be tomorrow and I'm going back to the West Coast and you will you you'll

find out where and why soon there's great news I'm super thrilled but this

This is the end of a five -year journey for me. And the beginning, hopefully, of another journey back to the West Coast.

Passion for Transformation and AI

So my main passion lies with transformation and technology adoption and how we can reinvent who we are, how we live, what we achieve with technology. And I've just been blown away this past year by AI.

And I've kind of embraced it entirely. There's not a single domain of my life which is not impacted by AI anymore. And I mean not a single domain, okay?

And I've let go of all the boundaries. But I would put everything on the cloud too. So I have to say I'm the kind of person who puts everything on the cloud to begin with.

And then I'm just my own experiment to some degree and everyone around me too.

Inside Moderna’s AI Journey

So a few words about Moderna.

Company-Wide Adoption

1We have 100 % adoption. There's not a single person in Moderna who doesn't use AI. At least the ones who have access to a device.

Scaling with Agents and Messages

and we we have built 10 ,000 agents which are in production which we just passed the mark of 10 ,000 which is pretty crazy for a 5 ,000 people company and we've

shared 10 million messages in the last five months this is on my list of 10 so it gives you an idea of and this is just getting started like this is not the end of the road but this is I would say a start in terms of adoption and then we

We have to wear the working, you know, there's a lot to be done still and proficiency, but in terms of adoption is there.

What We’ve Learned and How We Share

So I want to share with you a few of the things that I've learned from a lot of people around me because the whole culture is on fire at Moderna on AI. And also that I love to share and hear your thoughts.

Audience Q&A and Live Discussion

Before I go there, I would love to have a quick round table on what brings you here. like what would you love to hear about what you'd have to see and what can i do for uh for you with the presentation i would love to hear a few sound bites from the audience maybe i can adapt yes okay

okay all right okay okay okay okay that i will cover to some degree because we're going to cover everything but thank you this is congratulations for your your initiative and your entrepreneurship, and thank you for sharing your desire for today's session.

Who else? Yes?

I'm at a biotech. I want to do everything that you're doing at Modela. Oh, nice.

I don't do much at Modela, but people around me do a lot, and I make the conditions for them to do a lot, but yeah.

Okay, we can, specifically, anything that comes to your mind that you wish you would know more about?

More of the data analytics, probably. Okay, data analytics. It won't be covered in this session, but I'm happy to discuss it, and we can talk a little

On Data Analytics and Tool Use

bit about code on the way because code is a good way to start with analytics and by the way until we had analytics in generative ai there was just uh no way to to touch numbers right like you remember the times when you would be like what is the 213 multiplied by 5 and like the bi would go like i don't know like 600 and we had no and because it was trying to guess the answer and

then this thing came which is called tool use in which is like wait this is math i need to call someone who can help me and does math and then calls a python library that actually does math does multiplication gives you the answer back uh tool use is huge uh the use cases we have at

modena which are millions tens of millions hundreds of millions of dollars value use cases we have some of those they are deeply leveraging tool use like the ai has full surface understanding understanding from the instructions in plain language.

It differentiates the deterministic tasks and the non -deterministic tasks.

So for instance, understanding a patient comment is not deterministic, but finding if they made that comment within a seven -day window after injection is deterministic.

So this is different. If it's deterministic to use, if it's non -deterministic, LLM, and then knows exactly how to operate

both to the effect that it can be very, very useful to complex sequences of rule -based tasks that are sometimes deterministic, sometimes non -deterministic.

My two examples are actually coordinates of clinical trials, but there's like dozens of others.

And that tool use topic is huge, right?

Like now with all the data connectors, now with all the application connectors and so on, AI is becoming more and more a single pane of glass to both interpretative, human -shaped language and understanding and reasoning and super deterministic, thorough analytics, instructions and so on, but that LLM cannot touch with the 10 feet pole, that has to trust to a deterministic

software that it calls for the occasion and then uh should sound right all right so smaller smaller part but like so we'll be talking a little bit about code showing a little bit about how to interact but this is just peeling a tight surface of like the depth of what this can do when it's

deployed at scale oh lots of questions all right there okay okay yeah and technical writing is such

Technical Writing as a Hybrid Use Case

a great space at the intersection of deterministic and non -deterministic tasks, because it's non -deterministic by a sense, but it has to follow very strict rules. So that's another great topic.

When you get out of a template and start going into things that are very fixed, there's a lot of experiments being done there, but it's a great use case.

Human Augmentation, Not Replacement

I think about now, all of the technical writing is done with AI at this point. It wasn't the case a year ago, but the last year has been a - Yeah, yeah, yeah, yeah, yeah, completely, yeah, completely, and to be clear, when I say all the technical writing is done with AI, it's not done by AI, it's done with AI, and I am very much, if you follow me on LinkedIn or anything, you'll read this again and again and again and again, a proponent of human augmentation with AI, not of human replacement with AI, right, I don't believe in it at all, not today, not tomorrow, never actually,

It doesn't matter if AI becomes extraordinary in every field. It's not the topic, actually. Because the better it gets, the better we get. And we will always be sitting on top of it.

And there's no limit for how good we can get. So I say this because other people think, oh, it's like machines have gotten better than humans

at doing fabric for the textile industry 500 years ago. Well, there are still designers. And there are still people who do crochet for fun.

There are still people who are amazing at inventing fabric and patterns and new materials. So humans are all over the place. Just don't weave fabric so much anymore, but thank God, it's an inhuman task.

Elevating Roles with AI

So all this to say, in almost every single task that we have augmented with AI, we have done the work through workshops, through intentional design and reorganization, to elevate the role of the people who are not supported. And it's transforming the whole company.

The example I gave a minute ago about how did someone declare what they are feeling and is it in a seven -day window, this is typically the size of medical monitoring. monitoring, it's crazy how poor it is compared to everything you could get out of real -world evidence on tens of thousands of patients experimenting with a new medication.

Just saying yes or no they had an adverse event in seven -day window is dramatically low stakes, right? If you now turn the monitors into investigators, it's all right. You have AI now. What more can you find?

And then you brainstorm with them of what more can they find, because only they know, by the way. I don't know, but they know. And then you start reinventing the role entirely in something that they like much better.

And they don't need to do the manual task. It's done, it's automated. It's actually batch processed. So they come up in the morning, it's all done. And they can do review by exception.

And then they get to work on the real work, the work that has never been done before. Like what is really happening?

So, okay, this is just, so we will be talking about it. But the intent is always the human part, right?

Like we inherit influence, like genres and styles and things like this. We provide intent, and then we delegate execution. And I think that part between influence and execution, the intent, is uniquely ours to own. And that's where it's really interesting.

And we can also discuss it in the demo. I'll take one or two more perspectives, and then I'll go to... Just about... Yeah? Okay.

Governance, Risk, and Practicality of Agents

Those are all fair questions and concerns.

1In my mind, 1AI agents are like Excel spreadsheets. They're like what? Excel spreadsheets.

I don't manage Excel spreadsheets. I don't care their quality. And people do what they do.

Just it happens that like maybe a dozen of them are significant to the company. I think we have the company's budget on Excel spreadsheets. So that one needs to be carefully taken care of.

I think we have some of our accrual calculations that are done with macros in another Excel spreadsheet. I think we have state of inventory for some of the manufacturing on Excel spreadsheets. There's like a few dozen maybe. are really important and need an owner, need a lifecycle management, need quality control, and need versioning control.

For the rest, there's probably hundreds of thousands of Excel spreadsheets at Moderna. No one knows. Maybe someone knows better. I don't even think that our head of Microsoft applications knows. You see what I'm saying? And no one should care too much about it.

We give people a form of power with applications, with Excel, Outlook, Word, and so on. And mostly, they find their way around it with good culture. And sometimes, these elevate to the point that you need governance, and then you should.

How We Monitor Popular Agents

So, at Modar now, we have an agent that looks at all the others. And as soon as there are more than 20 people using an agent, it's flagged for us. So, we set the bar at 20, right? As soon as something is a bit popular, you quickly spring well above 20, right?

Like the latest agent we launched is a growth agent for people to think about their growth plan in their career. It has 76 % adoption, so it's something to the effect of more than 4 ,000 users. So you immediately see the outliers, right?

But it's less than 1%, like 50 of our agents do more than 50 % of the traffic of all the agents, and remember, there's 10 ,000. so this might become a more acute challenge in the future but we'll get better at it too

for now it's I've had to reassure my own leadership team that it's okay that people do small agents that just help them and they're the only ones using it or like a couple of people around them in their team or in their office don't breathe down their neck let them just use

this because we don't uh yeah it's it's fairly safe they don't have a right or act access to to systems in the company they have read access to some of the things it's very we're very early right like so i hope that it answers the question it's all a very fair question and i think our mental model needs to adjust to the variety of agents they are not that big of

a deal really uh yeah but that it would need it would need right access but but it would need right access but our platform doesn't have right access right but people are creating things where they can't like we prevent at the admin level right access so it cannot write or delete or erase

Security, Privacy, and Compliance

you can only read what's gonna happen yeah what so just to make it clear we have chat gpt enterprise we have zero data retention on all the apis we have uh i won't bore you with the acronyms of the standards of compliance we have all the centers of compliance you can possibly think of And we have sealed every possible entry into the logs.

There were six ways to access the logs. We documented them and we sealed them all. Why? Because our CEO uses it. Our CFO uses it. Our chief legal officer uses it.

And then the very first agent we launched was for the benefits in 2023. And people put private health information when they ask about benefits. So we, but we never, like, I think we started in April 23. as soon as GPT -4 launched with the API.

And a week later, we had sealed the six entries. And we told people, the first message they heard about us is, hey, there's this new thing. What can you do with it? And by the way, your conversations are entirely sealed and private.

With the limitation that we're a discoverable company, so the government has the right to request information from us, and that we can also be subpoenaed for information in case of litigation. So people should understand this is pretty much like their emails. else, but it has the same level of privacy.

And at this point in three years, we have never had to unseal anything. It's never happened yet. So, but privacy is very, very important. Otherwise, there's no momentum.

I'm just, I'm so, I'm so sorry to interrupt. And I'm, I'm, I'm, let me be, apologize in advance. You need to jump into demo? I'll jump into the presentation. Okay.

On ROI: Utility, Not a Project

I was listening to AI ROI sessions. Oh yeah. And I mentioned you guys to say that you don't have a policy internally that you have to justify.

No, yeah. ROI on, you know. Yes, no, no, my CEO forbids me from doing it. My CEO has, so in the early days, our CEO was always in favor. He's a pure geek, he loves AI, he believes in it, and so on.

But in the early days, the CFO was like, can you show me some of what you're doing, like the value? So it was the early days, it was like September 23, right? Like three, four months after we launched.

So I went onto the forum, we have 3 ,000 people in an AI forum on Teams. And I said, who wants to brag about what they're doing and what they're achieving? So I got a list of 100 people bragging. I took 20 of them, which made sense. We made a slide with it, so we have a slide with 20.

And each of them had calculated more value than the full cost of AI for the company on an annual basis, each of them. And we had 20 of them. So I was like, huh. So I presented the slide to the EC in a quarterly business review.

and then our CEO laughed and said I never want you to spend another second of your life calculating ROI on AI period and I was like alright I offered to him again last year because of course the stakes are rising a little bit it's we have connectors investments and so on and he said again I can't I need I need to explain to you again I never want to hear you have lost your time on the ROI

of AI so now I want to explain the one thing this is for day -to -day AI now some In some cases, we have huge investments in digital solutions, and they have an AI component. But it's not the array of AI, it's the array of the full digital solution. And actually, most of the cost is on data, right? So it's not on me, it's on someone else. We have a great relationship, but I don't own the data here. And then there's a lot of software programming.

So if there's a multimillion dollar project on a digital stack that includes AI, of course we should do roi of course but it's not an ai topic anymore they would have been all right anyway with or without ai right it's uh this is not an ai topic hope that answers your your

AI as Infrastructure

questions all right so the rest of ai i think of as utility like the internet electricity we are surrounded by electricity here like it powers the laptops the lights you know our phones and so on the aerial electricity anyone all right so we don't know we don't calculate.

And I think my CEO would fall from his chair if he learned that someone was spending time calculating the RA electricity at Moderna.

Live Demo: From Image to Interactive App

So what I'm going to do is I'm going to, so first, people hate demos in tech and I'm no different because they can go any direction. We just improve them as we go.

Today is particular because we have a new model, 5 .2, which launched at 1 p .m. today.

So I'm completely, yes? Sure. Louder, you mean? Yes. Okay.

And so I'm going

So what I'm going to do is I'm going to upload an image. So let me show you the image first. Actually, I'm going to show you from my download. It's going to be easier.

Introducing “Fred Habits”

So it's an image of what we call the Fred Habits. So what are Fred Habits? OK.

Fred, this is more like a Fred used in the 90s to ask his assistant to print the emails. And he would write his answers, and he would give them back to her so that she could type them into the machine all right so and you laugh

and you should like i laugh too but bear in mind that at the time you had to learn a laptop to plug in to know the os to boot it with dust to to put your software to understand and then there's email internet there was a little bit of friction you had to learn a little bit like now we do this in our sleep but it wasn't so obvious and so fred didn't want the little friction of learning he

just thought his assistant could do it all and by the way what's the laptop it's like a better type writer and it's not for him and of course now people laugh at this but we still have Fred behaviors just Fred behaviors of today right I'll give you

Modern Examples of Outdated Habits

one let's stay with emails no you're gonna laugh again some people open open their inbox and open their emails one by one I know it sounds crazy I know why would you do this when you can just use your connector to your inbox and ask a

GPT, what's in my inbox, prioritize and surface the most important emails, tell me what I can archive, and draft the answers for me on the top 10 of them. And it takes three minutes.

Actually, you should use a task that is ready for you at 8 a .m. when you wake up. But still, some people open their inbox and read their emails one by one.

And maybe you know this, the volume of emails is like 20 % or 30 % higher every year. It's one of those exponential lows. No one is on top of their inbox anymore nowadays. It's impossible.

And I can say the same about files in your folders, about threads in your messaging application, and so on and so on. Like, these are thread habits.

Opening a drive to find files, opening an inbox to find emails, or a messaging app to find messages is downright crazy.

That's becoming a little offensive. I'm trying. You see what I'm doing, right? But hey, it's downright crazy. I can't put it any better.

I remember the time when I used to do it, and I'm so grateful that I never do it anymore. Don't worry.

AI is not going to forget an email. It's not going to hallucinate an email. Those data connectors are really solid.

But you're going to start briefed and not overwhelmed. AI is going to brief you about the current state of your inbox, and you're going to make decisions from there.

so that's not what we're going to see though we're going to talk about all of the thread habits so that was one of them

scroll threads to catch up and open folders to find files and I think I have somewhere the email topic meh, there was

scroll threads to catch up but all of those are like bad habits right

improvised conversation meet with that transcript things we shouldn't do at all anymore right and so

Turning Habits into an App with Canvas

you might ask me what are the good habits And I have ideas, but let's find out what AI thinks about it.

So what I'm going to do, because this is still a fixed image, and I did that image with AI a while back, like maybe six months ago on an old version of ChatGP.

So I'm just going to add that photo. So let's find it here, the Fred Habits. And what I'm going to do, because I want to ideate a little bit, I'm going to trigger something called Canvas.

So if you're familiar, your Canvas is going to open a site document. It's great for ideating, it's great for preparing messages, or emails, or drafts, or documents.

It's also great for code. So by clicking Canvas, I make sure that it's not going to answer to me in the conversation, but create a document with whatever I ask next.

With it, I'm going to dictate because I don't, so I speak much faster than I type, but I read much faster than I listen. So my normal mode with AI is voice in, text out. I dictate and I read. And this is how I interact with AI almost all the time, unless I'm in a room and there's people, but today's going to work.

Okay, and so let's try this. Hey, so I'm in a room for an AI meetup with a lot of amazing, friendly people, and I want to show them something that we're going to do together.

I would like you to take the image that I just uploaded and attach to this presentation, to this request, and I want you to turn this into an application using Canvas. And it's very important for me that you use Canvas because then I might go into the code and do things there. So do this in a Canvas document.

I would like you to turn all of those eight tiles where I put super bad habits, like pre -AI habits, I want you to turn those eight tiles into flip cards.

And when we flip the cards, I want you to show what is the good habit in the edge of AI and ideally also give me an example of a prompt that I could use to enact that habit on a daily basis.

Because we're doing this I want to gamify a little bit. Give me something to the effect of a progress bar that if I open the cards I see some progress like we gamify this a little bit.

You can put a little bit of text on top to introduce the whole system and a little bit of text below to give me ideas of how to turn this into a good life habit and use your creative

Oh, one more thing. I would love the illustrations on the cards, at least on the front side of the flip cards, to be emojis. This way I don't have to deal with messed up illustrations or anything. It goes from the database of emojis. So find the best emoji for each front card. I trust you.

And blow our mind because I'm doing this in live presentation with 200 people, so I want you to show up your best foot forward, okay? Thanks.

So why do I say thanks? because I'm like this, right? Because I don't want to break character. If I start, like, bullying AI, I'm gonna start bullying everyone, and I don't want to become that person. Okay.

And let's go. Let's see what happens. This is gonna take... I don't know how long it's gonna take because it's my first time doing this with 5 .2.

You see the model over there? It's the new model from OpenAI that will launch today. And so the first thing, by the way, when I dictate, I read, and then I send. In this case, I was pretty sure it was gonna work fine. Let me put this a little bit bigger. So this is what I just said, right, like this is, yeah.

And so, and as you can see, the canvas has started, it's coding. So we let the thing code, do its thing, oh nice, I'm starting to see the emoji, do you see the emoji coming here? This is going to be fun, I can't wait to see what it does.

Editing and Iterating in Canvas

And while we do this, I want to say the canvas is editable two ways. You can open it and type it in the canvas, in the code. That's if you know code.

But even if you don't know code, let's say you're looking for a word, let's say you have the word scroll and you want to change it, you do control F, find scroll, and it's going to find the word in your code and you can just type the new sentence you want there. there, right, because the script is also encoded. You don't need to touch the structure or anything. You can just change the content.

So there's a number of ways that you can just improve with the conversation. We're going to do it together or improve in the code itself, either words or colors or parts or just. So this is very interactive. And once you get the hang of it, it's very playful.

From Prototype to Product

And there's two ways it's useful. The first way, it makes for nice prototypes. You can use a prototype of an app like this, show it to your users, and then they react. And so it's so much more than Figma. And at some point, you're going to have something pretty stable.

And you'll be like, all right, let's make a real app out of this, because this is still small apps. They run locally, but they're still small apps. And of course, the other possibility is that you use this as your final product.

So for instance, at Moderna, we have a lot of people who dashboard their data. So they import CSV files on the locally run dashboard. We see we can share it with the URL. And they don't use Slides or PowerPoint at all anymore.

Now, they just upload in live dashboards for leadership in weekly, monthly reviews, and they use those apps to surface data and to show visual intelligence on the surface of their data.

Demo Results and Iterations

So, it's done. 300 lines. That's a short code. Last time I did it, it was like 500. Let's see what it gives. Okay.

so uh which by the way um if i share it here i can create a link you're gonna see why uh the link is copied and i can just put the link here we can come back here later but uh it's gonna put it in full screen it's just better okay so eight byte habits to lose in 2025

it's a bit dark so maybe we can change the color and make it look better give me two seconds hey it's a little bit dark can you do like a beautiful orange to light blue gradient keep the glass paint i love it with the reflections but do something a bit more lively and entertaining and put the emojis in the center i want them front and center big emojis and small text thanks so i need to to send this and what it works we're going to discuss so here we are

with our app as you see i have my progress bar here which i asked for my flip card so of course the moment of truth is i'm going to flip a card so we had here a scroll twice to catch up what did it say oh i have a little bit of format issue here we can deal with it later ask for a crit digest plus actions turn noisy streams into a short brief what change what matters what to do

next yeah this is exactly what i meant all right so good find there another bad habit is keyword search the web right so no ask for answers with sources plus level of certainty this is a pretty good uh habit shift from hunting links to getting a source structured answer this is perfect

and then i'll do one last one but habits meet with a transcript my progress uh no i don't make progress all right this is not working oh yes it is working turn conversations into reusable assets capture the meeting once then reuse it forever summary decisions risk follow -ups and so on right

so we're starting to see the being of the flip cards i like how they light up though when i put my cursor and i didn't ask for this right like i voluntarily give my intent provide a few boundaries and then give room for the ai to demonstrate initiative in execution like the more i say at this point the less useful it is so i didn't have to say i want the cards to light up when i hover them with the mouse because i didn't even think of it and i think it's a and and I'm appreciating it.

And let's see what it put in the text below. Pick one card, make prompt reusable, body system, pair up. So little things, right, it's just getting started. And from there, we could do really literally anything. So I'm gonna go back to the first part.

You know, oh, this is very light. Okay, it went a bit far too much on the opposite. So there's a good balance probably at some point can be struck between the gradients. I liked, yeah, I think the other color scheme was better. But yeah, we can iterate and find and explore. The emojis are a little bit bigger.

So we could have style, we could have content, we could have features. But I think this is pretty much my demonstration there.

Publishing and Sharing Options

The last thing I want to say is that if you download this and you put this on any URL, you make it a full stack app. So ChatGPT right now doesn't have a publication component. It's still, you can share it with a link with anyone, but you can't surface it on the web.

however I'm going to do a danger of how it's where I put it I saw I had to create a domain for this and this is where I wanted to land with my beautiful gradient of orange and blue and with my big emojis and meet without transcripts give this and you remember we had this little format issue so I solved it by creating a show and hide button at the bottom at the bottom the last time I did this and I can hide all answers I can reveal all answers the the little features i did this with a i used lovable just copy and paste the code and then put it on the url it's a small ai app that creates a easy site to put forward and this hopefully talks also to your point about curriculum because it's much more interactive and and fun and you can shuffle the cards you can repurpose them one thing we can do here is uh let's see how that works

Pushing Creative Changes Live

so i'm not happy at all with the new colors i can't read white on light orange please do a gradient of orange and blue that's darker so that the white font shows up better in front you didn't put the emoji at the center of the card so i'm very sad about it like please put the emojis at the center of the card not on the upper left corner and i want you to change entirely the whole thing and to make it eight flip cards on the eight cardinal scenes and now you're going to tell me there are only seven cardinal scenes i know so i want you to invent an eighth cardinal scene that meets the moment of our society and on the flip side of the card i want you to show me the virtuous uh behavior that meets the cardinal scene that you just invented for all of them by the way i want the eight cardinal scenes to have a flip side in which I have eight cardinal virtues or something like this.

Thanks. Yeah, I know I say thanks, but... All right.

So then again, we're letting it work. I'll take a few questions while it works. Oh, I need to push the... I like that it always shows what you just dictated so I can read it and correct it.

In things like this, I'm not very worried. But sometimes if I see names or if there's specific elements, I want to just go back and correct the names and so on. Yeah.

Audience Questions on the Demo Stack

Isn't it tremendous? It's tremendous.

So one, just kind of a mechanical question. So you use ChatGPT to create the canvas. That creates the code. And then you pasted that into Lovable. And then Lovable just takes it and makes it.

Yeah, so you have options there. You have options there.

So we're getting into the weeds. I don't want to get too much into the weeds.

Languages, Tools, and Lovable

But there are a few languages that ChatGPT is really fluent with. So HTML is one of them. React is another. There are simple languages that you can just copy, paste, you're done.

All right, period. However, when you're in lovable, it has way more languages available and can do more bells and whistles and things like this

So you can either you need to instruct chat GPT to do pure HTML or pure react if you want to copy paste and launch Otherwise going to mix both and it becomes complicated

and You can either download in lovable and publish or you can download lovable say now use all the breadth of your languages But make this a good app and let's talk about the features and then and then you just ask for features

just ask there and you can add as many features as you want and then sky is the limit of what

you can do on on unlovable yeah so the the sister site of danger of of how is power of y it's the danger of how power of y um pro why also deal with lovable and don't need to go there

The “Power of Why” Agent

and it has 2000 pages of all my transcripts interviews uh book excerpts harvard business cases and so on you can ask any question on AI or AI transformation and it will answer in my voice with my examples from my 2000 pages of transcripts calling an agent in this case one of the latest agents and it does a pretty awesome job it takes like 8 to 10 seconds

so let's say I'm going to take how can I use AI in instruction design and learning I was going to take 8 to 10 seconds and start proposing things.

So I just want to say, and by the way, if you would scroll down, this is all the actual podcasts that it's taking from, the articles, the publications of them have their own page, so you can find the source if you want to.

And so, and there you go. So, and it, it, it normally is pretty good answers.

Like, at this point, it's pretty much I've, I've done hundreds of, of trial runs and I've had friends who've tried also. I'm usually pretty happy with what comes back from it. and say, well, I could completely say this kind of thing.

It used to have my voice on it with the voice clone, but it cost a lot of money, so I took it away. It's too expensive.

Yeah? Sure, this is a... I was hoping I'd figure it out eventually.

Using Personal Corpora and Vector Stores

So this is... We're talking about what we can do, how far we can push code.

And here I'm showing a case in which there's an agent that has all of the transcripts of all my public speaking engagements, 2 ,000 pages.

and that draws from them to answer any question on AI or AI transformation and I'm giving an example and that's the example I asked how can I use AI in instruction design and learning which was the question of one of our

participants at the beginning of the session and it says using AI in instruction design and learning is about more than just automating tasks which is so right it's about fundamentally enhancing how people learn how educators communicators create learning experiences, which is so me to say this.

From what I've seen, AI can serve multiple roles, and each has its own potential to transform instructional design. My voice on top of this, no?

And so first think of AI as a coach or expert that can help learners by providing precise feedback. This is, because thousands and thousands of pages of me speaking about those topics, this is really me.

Like I use it to pick my own brain, because I can't remember everything that I know. Like, you know when you're in the Uber and you think, God damn, I should have spoken about this at the iMeetup?

Like, we are not very good at picking up our own minds. So when someone asks me a question and I'm quiet, I go to Power of Why, I type the question, I read my own answer, I'm like, ah, that's a great answer. I love it. This is so me.

The only thing I'll say is that I always try to then, then AI is not yet very good at skating where the puck will be. So it's based on my past podcast interviews. But the things that I've experienced this week or last week are not there.

And the way that I'm starting to think about the future is not there. But it's a good baseline. I can start with this and then think, huh, and what have I been thinking those days or weeks that I could add to this and kind of freshen it up a little bit?

it no so when you when you work there's two ways for all those companies one way is the client so chat gpt and chat gpt enterprise and one way is the platform so if you type platform .openair .com you're with the apis right and it started as an api for a while i was using only the api the client came pretty late actually uh chat gpt so we had to have our own client it was called mchat at

Moderna is there was no client and so I used the API to do this and there's a vector store it's called like this so you upload all your documents into the vector store I can show you what it looks like well in that case and my

personal access but a model we have a github we have an enterprise account on github it's all our code bases on github and I'm gonna state we're gonna go in the weeds again and I'm so sorry for you who are not technical and then I said by right remember I'm not technical either a business school and consumer

brands. But I had to kind of learn a few of the tricks.

GitHub has its own security features. It doesn't matter if the code was written by a human or by an AI because both of them can be really bad at it.

There are some really bad human coders and there are some really bad AI coders.

No, no, but they are in the vector store. So technically speaking, so technically speaking they are in Azure open ai so that's a part of azure the cloud by microsoft that hosts the vectorized content

Why Enterprise Examples Aren’t Shown Publicly

that is propagated into the ai of openai because that's what i'm showing here right like when i'm in an event like this i show you personal stuff because i am not at liberty to show you enterprise stuff if i would do this i would create an immediate situation with insider training opportunities right because you would know i'm a perpetual insider i'm a perpetual insider at mora So I am not at liberty to share enterprise use cases in a public setting.

However, if this was in Investor Day or R &D Day or AI Day, I would love to answer all of your questions and show you the industrial -grade version of this, which is pretty awesome.

Data Stewardship Is Bigger Than AI

And AI is not the problem either for document storage or for code storage. It's a problem way before AI. It would be a problem long after AI.

AI is just a user has a dependency on good data, but it's not the topic of good data or good data management and good data retention, right?

Yes?

Leaving Moderna: What Happens to the Chats?

All right? And you're leaving Moderna. Yeah. Yeah.

You know, I love this question because I've been thinking about it a lot because I have more than 15 ,000 conversations in ChatGPT, Enterprise, and Moderna.

That's what I was going to get at. So sadly, they're going to be lost for everyone because they are private and sealed to me.

If ever we're in litigation, Moderna can open them. It's complicated to open them, but they can do it technically. and then use them in litigation.

I hope we will never get there. I've done everything I could so we'll never get there. But they are lost to me too.

And one of my propositions to OpenAI when I saw the situation, what was going to happen, is I told them I wish this was like electronic health records and that I could carry it forward from one company to the other without the confidential part.

And that my own thoughts and my own private ideas and even though it's complicated because the companies want you to believe that everything you did when you were there belongs to them, but the truth is a lot of it belongs to you.

I wish the AI was able to distinguish the part that is IP or proprietary data, take it out, and give me back like the 12 ,000 or something conversations that were not IP protected and that I wish I could carry forward.

And if I'm honest, because I saw it coming, I've been using my personal AI a lot during the last five years. Yeah, because, but not for work. But as soon as something is a personal thought of mine, on AI, on my usage or things like this, I use my personal AI.

So it's like having a personal mobile and a work mobile. You see what I'm saying? Or a personal laptop and a work laptop.

So when are you going to start wearing... No, no, it's a great question.

Well, I have a whoop band, which you could say is an AI device, but that's not... I'm just like Eric, I also have a phone that has instant messaging. I reorganized my phone entirely for it.

The Future of Devices and Trust

For now, we have our phones, the best we have. But I can't wait for the devices from OpenAI to launch. They keep saying that their main competitor is Apple.

Because they bought for $7 billion, the entire team which produced the iPhone. I think it goes without saying that their goal is to reinvent invent the way we have mobile intelligence with us.

That it is our main objective. Anything else is secondary to them, really. And Sam Altman has spoken about this time and again. So you would find this, you have to follow the breadcrumbs.

On-Device Models and Privacy

But this will run local models, so pure privacy. The models will never go to the cloud. They will be local on the device.

That's why you have all those nano and mini versions that they keep pushing, because they want to make sure the local models are really good.

and I'm very very eager to put my hands on the prototypes yeah for now I do with the mobile phone because there's there's no other way so we have to to reinvent buttons and to click on them and so this is so bad this is it's there there are

wearables but I don't trust them in the way they are built is not working first they use the cloud and use connectivity so there's a the privacy is not there at You don't want to speak everything also, right?

Yes, and you shouldn't need, but you see it's a broader conversation. But I love that we're having this conversation. We come from a place of knowledge and trust.

I can't wait.

From Village Trust to Global Tech

So I come from a village in France called Lombaise. You can't know it, it's like 3 ,000 people. But in that village where my parents met and my grandparents met, so I have four houses there that I call home.

In that village, I know everyone, right? Right, and it happened to me that the tailor of the village said, how Bryce, I made a suit for you for your hiring interview. And I was like, how do you know? And he said, well, what do you think, you know? And so, all right, so show me the suit, and show me the suit. I'm like, oh, this is really nice, I can't wait to wear it.

This is huge knowledge, because everyone knows everything. It's huge trust, because it's a village, literally. So maybe I'm the last person on earth who have a village, but I have a village.

And, however, when I come to the Silicon Valley, it's another world. This is not a village, and there's no trust. Like, I don't know anyone, and everyone has their own interest and everything. But the knowledge is being recreated.

It was already recreated by Google and by the search graph and by everything you type is remembered and all this. So the knowledge is uncanny. But the trust has not been restored. I don't know that I want these companies to have so much knowledge about me. And that's a lot of knowledge that's out there in the open air, right? And I don't have the trust to go with it.

Toward Absolute Trust for Absolute Knowledge

Now, if we project ourselves in a moment in which we have sensors that have health data, environment data, everything we say, everything we see, everything we feel, the temperature of our skin, you know, based on the hydration of our skin, our level of excitement, our angst, you know, You know, that's more data than anyone has. I don't even have that data myself. It's going to reveal things to me.

And the level of trust it requires is almost absolute. I have to have absolute trust in something that will have absolute knowledge. And I think that's what Johnny Ive is focusing on right now. Right?

The iPhone was the first version. a lot of of knowledge and actually a pretty good trust platform that people have trusted their iPhone and the stakes are rising on both sides there's gonna be more knowledge there needs to be more trust so all the I know the things you're referring to right like the the pins and all this like but they are like they're like gadgets I don't I don't trust them at all I don't judge them

Conclusion and Thanks

Bryce, thank you so much. Yes, thank you.