- il y a 5 semaines
Scaling with Vision: Resilience, Reinvention, and the Future of AI
Catégorie
🤖
TechnologieTranscription
00:00Hi, hi everyone, really excited to be here, so as you just understood, it's just going to be me unfortunately,
00:08I'm sorry, we had a little bit of a mishap and Fiji was supposed to be here with me on
00:12stage today, unfortunately she couldn't make it today in person, still going to be great, don't worry about it, we're
00:19going to talk about a lot of things, we're going to talk about retail, we're going to talk about AI,
00:24lots of really interesting topics, so Fiji, can you hear me?
00:30I can, hi.
00:31Hi Fiji, how are you?
00:33Very good, great to be with you guys.
00:35It's great to have you, sorry we can't have you in person, but it's still great to have you here.
00:41Sorry about that, but happy to be here.
00:44Yeah, awesome.
00:45So yeah, I'll try to introduce you real quick, just a word first for this, apologizing for this very weird
00:52thing, Fiji you are French, I'm French as well, and still we're going to do this interview in English because
00:57of course we have to think of the international audience here at VivaTech, so sorry for the French speaking people.
01:03But still, so yeah, I'll try to introduce you real quick, which is quite difficult because in 30 seconds you've
01:10made such an incredible journey, I mean you're arguably the most powerful French woman in tech right now.
01:18You've made a journey from CET, where you were born, the daughter and granddaughter of fishermen, to HEC Paris, to
01:26the US, where you've worked for basically all the major players of the Silicon Valley, you've worked for eBay, you've
01:31worked for Facebook with Mark Zuckerberg.
01:34Instacart, obviously, a company which you transform deeply, we're going to talk about it in a minute, and OpenAI in
01:42a couple months as number two and CEO of applications.
01:46So yeah, that's my first question, plus on top of that you're on the board of Shopify, you've been on
01:52Fortune's 40 under 40 list, etc., etc.
01:55Are you secretly powered by GPT-5 or something?
01:58How do you manage all that?
02:00Not that I know of, if I'm powered by anything, I think I'm powered by Nutella, that's what you keep
02:06going.
02:07Yeah, me too, but we don't have the same results, so obviously I'm doing something wrong, I don't know what.
02:11But yeah, that's a great power, that's a great energy.
02:15So yeah, this is your first interview since it was announced that you'll be joining OpenAI as CEO of applications.
02:21So it's great because we can have both a look at the rearview mirror, what you've achieved at Instacart, and
02:27look a little bit ahead at what you're going to do.
02:29I know you can't talk too, too much about OpenAI since you're not there yet, so for legal reasons we
02:34have to be a bit careful, but we're still going to talk about artificial intelligence, obviously.
02:38But let's start with Instacart.
02:40So Instacart, it's a very interesting example.
02:43Not all French people know it because it doesn't really exist here.
02:46It used to be a food delivery company, and you completely transformed it into basically a tech company.
02:53So you changed the scale and the nature of the company in a couple of years.
02:57Can you take us through this journey?
02:59What was your vision when you took on as a CEO in 2021, I think?
03:04Yeah.
03:05So when I joined in 2021, I had had the luxury of being on the board for six months and
03:10really seeing the company.
03:11And at the time, there were a lot of questions because it was right after the peak pandemic, and a
03:18lot of people were wondering if grocery delivery was going to be just a fad or something really enduring.
03:23And what I was seeing is not only the fact that this level of convenience was something that customers were
03:30going to want for a long time, but also the fact that there was an even bigger role for the
03:35company to play in helping the entire industry with our digital transformation.
03:39We heard a little bit about that in the session before, where the digital transformation of grocery and retail, which
03:47is by the way the largest category of commerce, has been happening both online and in store.
03:54And all of these retailers needed a trusted technology partner to help them with that transformation.
03:59So we set out to really build all of the technologies that power grocery transactions, not just online, but also
04:07across the store, and really help our grocers with that shift.
04:12On top of that, the other transformation we made is instead of just relying on deliveries for generating revenue, we
04:21realized that there was a very big opportunity with retail media and advertising and built an advertising platform by which
04:27brands can reach these customers.
04:29Which really helped us get profitable, be able to take this company public, but much more importantly, help our brands
04:37and our grocers actually succeed online.
04:40So what I'm really interested in is the tech part. You've put tech everywhere at Instacart.
04:46And as you said, you were trying to envision basically the shopping experience of the future and the supermarket of
04:51the future.
04:52And one really good example of that is you've implemented what you call caper carts.
04:57So it's basically the shopping cart of the future.
05:01I'm really interested in that. Can you explain us what they are, how they work?
05:05I think we have a video and we're going to see what they look like.
05:09Yeah, absolutely. And hopefully the video is playing in the background.
05:13But caper carts are essentially, you know, your usual shopping carts, except that they're massively enhanced by AI and computer
05:21vision.
05:21So that means that every time you drop products in the cart, the cart automatically detects what the product is
05:28and starts showing you on the screen what your total is, what the product's about.
05:36And then once you have that in front of you, you can do a lot of really interesting experiences, like
05:41showing you recommendations of other products that could complement what's in your cart, showing you where to get them because
05:48it's location aware.
05:49So you can go to the particular aisle to find that particular product, showing you particular discounts.
05:54I think you're seeing one on the screen that's allowing you to save money, which you may not have found
06:00on your own.
06:01We have also rolled out the ability to do gamification, where you can go on quests and scan a variety
06:09of products to be able to save.
06:11And as anyone who has ever shopped with kids in the grocery store knows, you know, it's good to have
06:18something to keep the kids entertained.
06:19And so now you have a platform with a screen in front of people that knows what's inside the cart,
06:26knows where you're at in the store, where you can really display a lot of, you know, enhancements to the
06:32product experience.
06:33And enhancements that exist online and everybody expects online, but don't really exist in store because stores are deeply unpersonalized
06:42right now.
06:42So that's why we think this is a winning model, because it doesn't change what people do.
06:47They don't have to learn a new technology.
06:49They just have to take the cart and like, you know, how they do all the time.
06:54But now there are these enhancements that allow them to have a richer experience when they're inside the store.
07:00So does this cart guide me inside the store like an indoors way?
07:04So if I'm at the cheese section and I want to go to the charcuterie section because I want to
07:09do a raclette, and yes, it's 30 degrees outside, but still, we're going to do a raclette, no problem.
07:15Always a raclette, yes.
07:16It does lead you to the aisle to find your complementary products.
07:21And it does guide you in general, you know, if there's a particular deal, it tells you, hey, if you
07:26go to this aisle, you're going to be able to unlock that deal.
07:28If you go to this aisle, you're going to have a complementary product.
07:32And that also becomes a platform for advertising, where advertisers can actually use that to promote certain products in the
07:39context already of where you're at in the store and what's in your cart.
07:42So it's a great transition to artificial intelligence because obviously there are sensors, cameras, AI in your caper carts.
07:50There's been a lot of talk and a lot of hype about completely autonomous stores.
07:54We've seen the Amazon experiment, Amazon Go, but they are not the only ones.
07:59This idea of a store with basically one customer and AI everywhere, algorithms, cameras and sensors, and everything is just
08:12easy for the customer.
08:13But it doesn't seem to be working that well in real life.
08:18So what do you think about that?
08:20And more generally, what do you think the balance should be between automation and human labor in retail and in
08:28general?
08:29Yeah, I think when we look at what customers want, they do not want a store that's like completely devoid
08:36of people.
08:37They actually want to go to a store where there's going to be some real human activity.
08:42And what we've seen so far is that these technologies can really help augment human labor.
08:48When we deploy caper in stores, we're now in 60 cities and a lot of the major retailers.
08:54What we're seeing is that it actually frees up in-store labor to do more customer support,
09:00to go to the person who's clearly looking for something and be like, hey, how can I help you?
09:05And these are ways to actually make it more enjoyable for customers to be inside the store
09:12because they have the help of the car, but they also have the help of human labor.
09:15And for, you know, the people working there to actually focus on higher value actions
09:21rather than, you know, just trying to scan what's on the shelves if that can be done automatically.
09:26So we think that it's really the magic combo and we see that as augmentation, not replacement.
09:32Okay, that's a fair point.
09:33And so there's a lot of talk, obviously, about generative AI as well.
09:37So as I said, we can't talk too, too much about open AI,
09:40but it's completely transforming the ways companies operate.
09:44Maybe you can talk a little bit about the way you've incorporated AI at Instacart, for instance.
09:51Are all Instacart employees augmented by AI?
09:54Are they all using ChatGPT on a daily basis?
09:58How do you make this work?
09:59Yeah, of course.
10:00So the way I'll talk about internally and then what we did in our products as well.
10:05So internally, the mandate that I gave the team is to implement AI, but implement it in a way that
10:12solves real problems.
10:13Because, you know, some of the challenges you see sometimes companies wanting to check the box on AI,
10:20but AI being kind of a solution in search of a problem.
10:24Instead, what we did is team by team, we ask every team leader,
10:28what are some of the things that are most tedious, most repetitive, consuming the most time in your organization,
10:35and how can they be transformed potentially by AI?
10:39And that really forced us to focus on the things that would have the most impact.
10:43So to give you a couple of examples, in engineering, we now have 87% of our code that is
10:49AI-assisted.
10:50That is a massive increase in velocity and productivity.
10:54In marketing, we took a library of images that we had shot of food,
11:00and now we're giving that library of content a second life by turning them into animated videos that can generate
11:06new campaigns.
11:07Again, something completely new.
11:09In communications, we had this great idea of creating personas for journalists,
11:18so like a business journalist, a tech journalist, through AI,
11:21and now we train our spokespeople with these journalists.
11:25In legal, we had a ton of recurring labor that was all about, like, finding specific terms in contracts.
11:33Now that's all done by AI.
11:34And so if you look across the board, yes, the entire company is AI-enabled,
11:39but in every team it looks different because we have really tried to focus on, like,
11:43how do we help each individual team with our particular challenges?
11:47So that's kind of on the internal side.
11:49I would say on the external side, the thing we focused a lot on is how can AI really help
11:55improve our customer experience,
11:57and the main thing was really personalizing the experience to the extreme.
12:02We have this thing that we call internally a personal planogram,
12:07which is the idea that when you go to Instacart, the set of products that you see should reflect all
12:13of your tastes,
12:13all of your dietary preferences, how many people are in your household.
12:19And with AI, we have been able to take, you know, 17 million products and catalog them a lot better
12:27so that we know which products are gluten-free, which products are low-carb, et cetera,
12:31and then match that with people's preferences so that you have the absolute most personalized experience,
12:37which is really the holy grail of personalization, which we couldn't do at all at this scale before Generative AI.
12:44Okay.
12:45Well, you seem very, very ready to work for OpenAI.
12:49So, yeah, you're going to work from August on for arguably the most exciting tech company in the world right
12:56now.
12:56So, you will be the CEO of Applications.
13:01What does that mean?
13:02What is the CEO of Applications, and what does it mean for you personally to be appointed at that role?
13:08Yeah.
13:09So, you know, I've been on the board of OpenAI for the last year,
13:13so I had, again, a front row seat to seeing the company operate,
13:16and I believe this is a company that's going to change every part of our life.
13:20So, it's a real privilege to be, you know, playing a role there.
13:24In terms of what I'm going to do specifically, my team is really in charge of taking the incredible research
13:30breakthroughs
13:31that the company generates and really making sure that we distribute them to society in the best possible way
13:39so that people see the benefits of AI and, you know, that gets reflected into an empowerment in their lives.
13:47I really believe that if we look to the future, AI is going to empower us in a variety of
13:52ways.
13:53It's going to allow us to reclaim some of our time through productivity.
13:56It's going to make us express our creative potential even more through all of the creative tools.
14:01It's going to allow us to get access to knowledge and wisdom that was in the past, you know,
14:08limited to the most fortunate among us.
14:10It's going to allow us to, like, gain power in many different ways.
14:17And the best technologies do that, right?
14:19They take things that are limited to, you know, a certain set of people and make it available to all.
14:25And I see my job as making sure that AI has, you know, greatest positive benefit for society
14:31and that these benefits are equally distributed.
14:34So you'll be reporting directly to Sam Altman.
14:38He seems quite excited to be working with you.
14:41When you were appointed as number two of OpenAI, he tweeted this.
14:46Let me find it.
14:47I cannot imagine a better new team member to help us scale the next 10x or 100x.
14:54Let's see.
14:55So it seems like you have your work cut out for you pretty much.
14:58What should you do to grow OpenAI 100 times?
15:02What are the next big steps, according to you?
15:05Yeah, I think if you look at the current states, like, you know,
15:10Chad GPT reaches hundreds of millions of people.
15:14But fundamentally, AI should change lives for everyone.
15:18So we need to be reaching billions.
15:21And then once you reach billions, you also need to make sure that each person can find value in multiple
15:28use cases.
15:29I was talking about the use cases in productivity, use cases in education, health care.
15:35Like, the list goes on.
15:36Every single person really should have, like, multiple agents and a team of agents at their service
15:44to be able to help them get life done in a way that's much easier than what we know today.
15:49And so when you think about that and you think about AI being really an ambient layer,
15:55not just delivered through Chad GPT, but potentially through devices,
15:58and, you know, we made the acquisition of I.O.,
16:01through a variety of devices, through a variety of services,
16:04you can start thinking that 100x is just the beginning.
16:08And there's definitely a lot of value to be added for, you know, every human on the planet.
16:14Wow. 100x is just the beginning.
16:16That's quite an ambitious statement.
16:18Awesome.
16:19Yeah, so you were talking about Gen.AI and the way companies try to implement it.
16:24I read several surveys showing that 80% of internal Gen.AI projects in companies, they just fail.
16:33I mean, I don't know why.
16:34Is there a secret recipe?
16:36What should companies do, business leaders do, to make them work beyond just a proof of concept?
16:43How can we make this work?
16:46You know, I think it goes back to what I said earlier.
16:49Like, when you implement AI just for AI's sake, like, you know, you're going to be, like, looking for a
16:55problem with this solution instead of really starting with, like, what are the problems in your company that you're trying
17:01to solve?
17:02And every company has a set of things that are repetitive, tedious for their workers.
17:09And if you're able to identify that and then go and find the AI solution that's going to be a
17:15solution to that problem, you have a much higher likelihood of finding value rather than telling your entire team, hey,
17:21everyone, go use ChatGPT.
17:23And, you know, no one knowing what to do with that.
17:25I also think that training is going to be really important.
17:28If you look at what we did at Instacart, we developed a set of trainings and a set of advocates.
17:33So, you know, we picked a couple of people who are very deep in AI.
17:36And in every team, there are the people that the rest of the company goes to for advice because that's
17:42not always completely intuitive.
17:44It's a very new thing and a very new way of working.
17:47So, you want to have the right training in place and the right advocates inside the company to help broad
17:53adoption.
17:54So, there is another huge field of exploration for artificial intelligence, and you've mentioned it earlier.
17:59It's healthcare, obviously, a topic you are particularly interested in.
18:03And can you tell us a little bit more about the work you've done in a personal capacity to bridge
18:08the gap between research and care using AI?
18:14Yeah.
18:14So, in 2020, I was diagnosed with a pretty nasty neuroimmune condition, and I went pretty deep on understanding how
18:22we're going to end up finding cures for these conditions.
18:25And the very good news is that the AI models are becoming so much more powerful that you could really
18:31imagine that with the amount of intelligence that's going to be generated, we can accelerate the path to cures.
18:38The bad news, though, is that if you don't have the right underlying data to really deploy these models on,
18:46that's going to be very hard to get there.
18:47And in a lot of conditions, we don't have the data.
18:50And so, I created a research institute that's really in charge of gathering all of the biological data, clinical data,
18:58genetics, et cetera, on patients longitudinally, so that once the model gets there in terms of intelligence, we're going to
19:05be able to unlock new insights that could hopefully lead to drugs.
19:11And then, on top of that, you know, we talk a lot about AI and healthcare in the context of
19:16finding new drugs, but I also think there's more immediate applications, which is right now, nine out of ten people
19:24say that they do not understand health information well enough to make informed decisions about their health.
19:31When you think about it, that's a mind-blowing stat, because the decisions you make about your health matter enormously.
19:37And with AI, we can really make that knowledge available to all and really rebalance things between the all-knowing
19:46doctor and the patients who just has to implement what the doctor says.
19:50We can actually, like, really inform the patients with what's happening with their health, not just when they go to
19:56see the doctor, because health care shouldn't happen at the doctor's office, but every day what they can do to
20:01slightly improve their health.
20:02And I think that's going to be a massively positive use case for AI.
20:07Awesome.
20:08So, we have ten seconds left.
20:10I have one last question.
20:12It might be a bit early for you to answer, since you're not already at OpenAI, but what's next?
20:19You've done eBay, Facebook, Instacart, now OpenAI.
20:23Do you see yourself running a French company, for instance, someday?
20:28You know, I'm just starting this new job, and as we've just discussed, it's a very, very big mission, so
20:35I'm going to be focused on it for a very long time.
20:38France always has my heart, but, you know, there's definitely a lot to do at OpenAI.
20:44Awesome.
20:44Thank you very much, Fiji.
20:46We can applaud Fiji Simo.
20:47So much, everyone.
20:48We hope you do the best for your new job at OpenAI, and we hope to see you in person
20:53at VivaTech next year.
20:54Thank you very much.
20:55Thank you.
20:56Thanks.
20:56Merci à tous.
20:57Bonne journée, bon salon.
21:00Bonne journée, bon salon.
21:00Thank you.
21:00Merci.
Commentaires