"I Want to Shatter Your Unrealistic Fantasies About AI and Rebuild a Proper Understanding" | A Conversation with Leding Li, Baidu Principal Architect in the Trenches

This is the most substantive episode of "Crossing" ever!

In this episode, guest Leding Li gives a thorough yet accessible introduction to generative AI, breaks down AI entrepreneurship directions, discusses the AI strategies of Silicon Valley giants, and comprehensively covers hot industry topics including Professor Yann LeCun, world models, and AI search. This is the 20th episode of "Crossing," and in my view, the most valuable one yet — full of common sense, fresh insights, and abundant inspiration.

——

Today's guest, Leding Li, is a treasured friend I've known for over ten years. We've road-tripped to Inner Mongolia together, eaten our way through Beijing's Japanese restaurants, chatted about everything under the sun, and witnessed each other's growth.

Mr. Li joined Baidu in 2010 as a Principal Architect, consistently taking on the most challenging assignments.

Today, there's broad consensus in the tech industry that generative AI will become the most important and largest-scale technology infrastructure of the future. Before generative AI, the two most important and largest-scale technology infrastructure buildouts in tech history were search and cloud services. In Mr. Li's 12-year career at Baidu, he spent the first six years as a search architect and the latter six as a cloud services architect. Architects who have experienced both of these massive infrastructure buildouts are rare indeed, globally speaking.

To be qualified for large-scale infrastructure architecture requires not only deep technical understanding but also intimate knowledge of every role and component in the system, in order to formulate systemic strategies and complete architectural design.

In today's AI field, opinions abound and many voices are strident. Notably, however, those speaking out most actively are mostly industry observers — particularly media professionals and investors. This is precisely why "Crossing" invited Mr. Li to share his perspective on generative AI's past, present, and future:

Only those in the trenches can hear the sound of gunfire; and only commanders and architects who have experienced battles of comparable scale can understand the deeper meaning within that gunfire.

——

🎈 Special thanks to Beijing Keyone Studio for providing the recording space.

——

🎬 Timeline:

Part 1 Shatter Those Unrealistic Fantasies About AI!

On AI: too many fantasies, too little rationality. Fantasies that aren't shattered are bubbles; shattered fantasies may become opportunities.

02:19 The layers of fantasy

04:42 Why so many fantasies?

05:55 How to shatter fantasies? Start by understanding how AI actually works

09:44 What harm comes from indulging in unrealistic fantasies?

10:50 On the AI bubble: it's inevitable and worth embracing; but also recognize that it really is a bubble

12:05 What has and hasn't changed in the 18 months since ChatGPT 3.0's release?

Part 2 Disrupt! Google

Mr. Li served as Chief Architect of Baidu's Open Platform (Aladdin), making him one of the most authoritative voices on AI search.

15:07 Will AI search disrupt Google?

16:45 What is Perplexity's value?

18:42 What is Google's Achilles' heel? This is where Perplexity's opportunity lies

19:47 In building RAG, whoever first understands that AI is the least important of the key elements may win.

Part 3 Take the Leap! AI Entrepreneurs

Understanding generative AI's technical principles and洞察ing user needs and scenario characteristics — what entrepreneurial choices emerge?

Wrong directions for AI entrepreneurship:

25:18 Direction 1: Using AI to plan travel itineraries; people's goals are often clarified through the process of searching and browsing materials

27:50 Direction 2: Using AI to fully manage ride-hailing, flight booking, and hotels

Scenarios suitable for AI entrepreneurship:

29:58 Characteristics: high margins + high fault tolerance

31:28 Low-hanging fruit: digital advertising, selling GPU cloud compute

33:01 Generative AI has made Douyin's recommendation algorithm even more powerful

37:01 Two bright spots seen in Chinese AI entrepreneurship projects

39:20 AI PaaS projects lacking distinctiveness

On Scaling Law:

40:49 The current state of affairs is the ceiling of Scaling Law; just because a car is a vehicle doesn't mean it will fly

41:57 Rapid progress we can expect from: model knowledge coverage, response speed

What highly certain developments will occur in the next 6-12 months?

43:04 Multimodality + mechanical control, sensors

44:25 On-device small models

44:55 AI is actually already everywhere

46:56 Advances in AI compilers will break Nvidia's monopoly

47:52 AI's integration with frontier scientific fields

Part 4 Defend! Silicon Valley Giants

The giants hold the highest form of industry moat: ecosystem; therefore their AI strategies are all defensive.

49:24 Meta: the entire AI industry owes Meta thanks

54:55 Microsoft: its rise to $3 trillion today is well-deserved

56:29 OpenAI has been boxed by Microsoft into the position of a model researcher

58:41 Predicting GPT-5

59:13 Apple: always experience-first, worthy of appreciation

01:01:31 Amazon: has presented an excellent three-layer AI system

01:03:39 Advice for Dify and LangChain: exceptionally impressive products, but if it were me, I'd advise them to sell to a major tech company as soon as possible

01:05:03 Tesla: FSD is directionally correct, but the road is long; Robotaxi isn't worth pricing in

Part 5

01:10:46 The AI strategies of domestic Chinese tech giants are all quite good — and that's not just being polite

01:12:16 Recommended AI information channels

01:20:19 If you could travel to the future, what answer would you most want to sneak a peek at?

01:22:30 What are the most worth-following issues over the next 6-12 months?

Part 6 YYDS! Professor Yann LeCun

The person on this planet with the deepest and most correct understanding of AI.

01:13:45 In the AI industry, what separates those who do well from those who do at the top level?

01:14:40 Who on this planet has the deepest and most correct understanding of AI?

01:14:54 Why are Yann LeCun's widely controversial views actually correct?

01:17:59 A primer: what is a "world model"?

——

🚦 Welcome to subscribe to "Crossing," where we follow the industry transformations and new entrepreneurial opportunities brought by the new wave of AI technology. "Crossing" is Steve Jobs' metaphor for Apple — standing at the intersection of technology and liberal arts, where great products are born. AI is transforming every industry; we seek out, interview, and bring together "active actors" of the AI era, exploring and embracing new changes and new possibilities alongside them.

👦🏻 Host Koji: Co-founder of The Fair and Tangdao, and someone actively diving into AI himself. [Koji's Jike profile], [A self-introduction from Koji]

👧🏻 Host Ronghui: Works at a tech VC, former Silicon Valley correspondent for Yicai. [Ronghui's Jike profile]

——

Welcome to join the Crossing membership community:

☀️ Daily AI Fresh News Brief

👫🏻 Encouraging everyone to date/make friends/find future collaborators

🦀 Add our assistant on WeChat to join: Rwkfbcianvd

——

💡 Inspiration posters

💡

💡

💡

——

🎄 This podcast is supported by The Fair's Sound Forest Podcast Project.