Baidu released in 2000 as a search engine platform. Two years lateron, it’s endedupbeing one of the coupleof business in the world that provides a complete AI stack. Its core companies period mobile, cloud, smart driving and other development efforts, and its items and services have drewin hundreds of millions of users and hundreds of thousands of business clients. Leading all of that is co-founder, CEO, and chairman Robin Li. He discusses how Baidu hasactually constructed generative AI into its organization – consistingof their AI chatbot, ERNIE Bot. Robin likewise shares the innovation patterns he’s keeping an eye on – from AI bubbles to robotaxis – and how he preparesfor these fast-moving modifications will change our world.
ALISON BEARD: Welcome to the HBR IdeaCast from Harvard Business Review. I’m Alison Beard.
Baidu introduced in 2000 as a search engine platform. Fast-forward 2 years, and it’s now one of the coupleof business in the world that uses a complete AI stack. Its core organizations period mobile, cloud, smart driving, and other development efforts. And its items and services have broughtin hundreds of millions of users and hundreds of thousands of business clients. Today’s visitor is leading all of that. For the 3rd episode in our unique series on the future of company, we’ll hear from Robin Li, the co-founder, chief executive, and chairman of Baidu. He discusses how his business hasactually constructed generative AI into its organization, the innovation patterns he’s keeping an eye on, and how he expects these tools will change our lives. Robin spoke to HBR editor-in-chief, Adi Ignatius, and took concerns from the audience throughout our current virtual Future of Business conference. Here’s their discussion.
ADI IGNATIUS: So Robin Li, I understand it is extremely late your time in Beijing, so we value your signingupwith us live for this, welcome.
ROBIN LI: Hi, Adi, thank you for having me. It’s fantastic to be here.
ADI IGNATIUS: Well, it’s terrific to have you. Before we begin, let me simply advise everybody in the audience to put any concerns you have for Robin in the ask the speaker chat, and I will attempt to get to as numerous as I can lateron. But Robin, let’s get to it. So Baidu, your business presented a ChatGPT-like item, ERNIE Bot, last year, that last I saw has more than 300 million users. I presume it’s been a knowing experience for you. Can you talk a little bit about what you’ve discovered giventhat the veryfirst variation came out and how it has progressed? Just inform us a little bit about all that.
ROBIN LI: Yeah, sure. We introduced ERNIE Bot, I believe, March 16th of last year. I believe that was the veryfirst ChatGPT-like chatbot for all the public business around the world. Because we’ve been investing in AI, particularly natural language-related AI for rather a coupleof years, we were able to rapidly launch a chatbot based on our big language designs. Over the past year and a half, a lot has tookplace. The innovation has progressed really rapidly and significantly. Things we foundout? There are a lot of things that I oughtto pointout. The veryfirst is that a lot of individuals, users, designers, the clients — they not just care about the effectiveness of the design, they likewise care about the reaction speed. They likewise care about the cost, reasoning expense. So after March of last year, we have rolled out a series of language designs or structure designs to please all kinds of various requires in various situations. Meaning that the design size might differ significantly and the reasoning expense might be really various too. And in specific cases, users puton’t mind to wait 10 seconds to get the finest address, and in other circumstances, you will have to do it extremely, extremely rapidly, sub-second action time. And likewise in terms of expense, we’ve been able to decrease the expense by about 99 percent, significance the existing reasoning expense is about one percent of the initial expense when we veryfirst introduced that. Having stated all of that, I would state that mostlikely the most substantial modification we’re seeing over the past 18 to 20 months is the precision of those responses from the big language designs. And I believe over the past 18 months, that issue has quite much been fixed. Meaning that when you talk to a chatbot, a frontier model-based chatbot, you can essentially trust the response. That’s a big distinction.
ADI IGNATIUS: Now, from my viewpoint, perhaps this is a UnitedStates viewpoint, there was a big wave of enjoyment about AI — especially with the release of generative AI items. You’ve talked about search, there puton’t appear to be a lot of or as numerous intriguing usage cases as perhaps some of us had anticipated by now. So I’m interested in your view, are we in an AI bubble at this point? What’s the trajectory of the innovation?
ROBIN LI: I think like numerous other innovation waves, bubble is kind of inescapable. When you pass the phase of preliminary enjoyment, individuals would be dissatisfied that the innovation doesn’t satisfy the high expectation created through the preliminary enjoyment. We’ve seen this lotsof times when the web took off in the mid to late ’90s, and there was a substantial bubble. For mobile internet, comparable things occurred. And this time for generative AI, I believe we will likewise go through that kind of duration too. But I believe it likewise assists, it will wash out a lot of those phony development or items that doesn’t have market fit. After that, mostlikely one percent of the business will stand out and endedupbeing substantial and will produce a lot of worth, will develop incredible worth for the individuals, for the society. And I believe we are simply going through this kind of procedure. This year the sector is mostlikely cooler than last year, however I believe it’s likewise muchhealthier than last year.
ADI IGNATIUS: And what’s the right company design? I indicate, there’s some big designs, Meta’s Llama, for example, are open source. Others are closed source like OpenAI’s GPT. Baidu I believe is promoted for a closed source technique. What’s the believing behind that and how does that set Baidu up to capitalize on AI innovation?
ROBIN LI: Yeah, you discussed closed source, however I would choose call it a commercial-grade design, structure design. I believe when you appearance at the most sophisticated language design or structure design, most of them are closed. And when individuals talk about open source, it’s kind of deceptive to me. It’s various from the open source of Linux or Python since for an open source design, what you generally get is a lot of criteria. You puton’t understand how those criteria was obtained and you have no method of altering those specifications. So it doesn’t have the impact of lotsof, lotsof individuals from various part of the world, and contribute back to the primary branch and make it muchbetter and muchbetter. Language designs are really various, and you can usage a so-called open source design to do things. But it’s really hard for you to contribute back. And another understanding individuals have on t