Alexandr Wang - A Primer on AI - [Invest Like the Best, EP. 272]

Here is the translation of your analysis chapter, following all specified rules.

At a Glance

Alexandr Wang (Founder & CEO of Scale AI) discussed the core building blocks of AI on a podcast, emphasizing that data is the "new code" for AI and pointing out that the lack of high-quality data infrastructure is the industry's bottleneck. Scale AI provides data solutions for Meta, Microsoft, OpenAI, the U.S. Air Force, and others, with a company valuation exceeding $70 billion. Wang argues that the investment logic for AI differs from that for software, as data quality is more important than algorithms. Drawing an analogy to AWS's cloud service model, he believes Scale AI will fill the "storage and computing" gap in the AI field.

AI Geopolitics: From Physical Battlefields to Digital Battlefields, Data and Compute Become the Core of the New Arms Race

Alexandr Wang believes that AI is fundamentally changing the nature of conflict, and future competition will primarily take place in the digital domain. Having grown up in Los Alamos, the birthplace of the atomic bomb, he argues that nuclear deterrence once brought long-term peace, while AI and cybersecurity are creating a new, more complex paradigm of deterrence and conflict.

• Paradigm Shift: Wang points out that conflict is shifting from "physical battlefields" to "digital battlefields." He predicts that 90% of future conflicts may be resolved or decided in the digital domain before they physically erupt. As a "cross-sectional technology," AI can permeate multiple levels, including intelligence, cyber warfare, and actual combat operations.
• Three Key Resources: Wang proposes that scaling AI capabilities depends on three core vectors: talent, compute, and data. He argues that the "power law effect" is stronger for AI than for software, where top talent (e.g., OpenAI's team of just 250 people) can have an outsized impact. Therefore, high-skilled immigration policy is crucial for the U.S. to maintain its leading position.
• Public-Private Partnership Gap: Wang notes a "cultural problem" in the U.S. regarding public-private partnerships in AI. Unlike Chinese companies like SenseTime and Face++, which directly serve the government, top U.S. tech firms have ambiguous and controversial relationships with government entities (e.g., the Department of Defense, intelligence agencies), as seen in the Google Project Maven incident. He believes that actions taken in the next 5-10 years will determine the landscape of great power competition.

Data is the New Code: The Differentiating Advantage in the AI Era Shifts from Algorithms to Data Assets

Wang presents his core thesis: "Data is the new code." In the AI era, a company's moat will shift from code to data. He elaborates on how this transformation reshapes business logic.

• Algorithm Commoditization: Wang points out that across different fields like image recognition and speech recognition, the top-performing algorithms actually use the same underlying codebase. This means the code itself has become commoditized and is no longer a competitive advantage.
• Data as a Strategic Asset: As the software component of AI rises from 0.01% to 50%, a company's source of differentiation will shift entirely to its data assets. A company's strategic advantage will depend on: 1) its existing data assets; 2) its "engine" for continuously producing differentiated data.
• Business Implications: Wang predicts that every industry will need to build its own "TikTok recommendation algorithm" for customer lifecycle management, business process automation, etc. This will lead to a significant reduction in marketing and sales expenses, as algorithms can predict a user's next action more accurately than marketing. He warns this could exacerbate inequality, as the technology itself "favors scale."

AI Capability Map: Can Automate "Digital Repetitive Labor," but Struggles to Replace "Physical World Tasks"

Wang clarifies a common misconception about AI capabilities: what is easy for humans is not necessarily easy for AI, and vice versa. He provides a framework based on "data availability" to understand the boundaries of AI capabilities.

• Core Framework: The problems AI can effectively solve depend on the existence of large amounts of easily accessible digital data. If such data exists (e.g., 20 years of Reddit data for training GPT-3), AI can perform exceptionally well; if it doesn't (e.g., data for a home robot folding laundry), progress is slow.
• Counterintuitive Examples: Wang notes that DeepMind and OpenAI recently released algorithms capable of solving highly difficult mathematical theorems and competitive programming problems. These are top intellectual challenges for humans, but because there is abundant, verifiable digital data, AI can actually perform very well on them.
• Impact on Employment: Wang believes that white-collar knowledge work (dealing with Word and Excel all day) is more susceptible to AI automation than blue-collar physical labor. He predicts AI will first automate "digital repetitive labor," freeing humans to engage in more creative work.

Position Moves

Position	Guest's Stance	Key Data
Scale AI	Bullish (own business)	Valuation over $70 billion; clients include Meta, Microsoft, OpenAI, U.S. Air Force, Toyota, GM, etc.
Tesla	Neutral (strategy mentioned)	Its autonomous driving strategy uses all Teslas as data collection vehicles to acquire diverse, rare-scenario data
OpenAI	Bullish (technical capability)	~250 employees; its GPT-3 model used 20 years of Reddit data; released algorithms capable of solving high-difficulty math problems
DeepMind	Bullish (technical capability)	Released AlphaCode, which beat the median human participant in competitive programming (i.e., outperforming 99.9% of humans)
Google	Neutral (mentioned)	Its Project Maven incident is a classic case of public-private partnership conflict; its data assets are a huge advantage
Meta	Neutral (client)	Client of Scale AI
Microsoft	Neutral (client)	Client of Scale AI
Amazon (AWS)	Bullish (business model)	Source of inspiration for Scale AI; its strategy of "parallel execution" and "building products from raw capabilities" is admired by Wang
Stripe	Neutral (analogy)	Its developer API and platform strategy are referenced by Scale AI
SenseTime / Face++	Risk warning	As contractors for the Chinese government, their AI applications (e.g., facial recognition) contrast with those in the U.S.

Judgments Worth Remembering

1. Data is the new code (Alexandr Wang): In the AI era, algorithmic code is becoming commoditized. A company's core moat and source of differentiation are its data assets and the "engine" for continuously producing high-quality data.

2. AI capability depends on data availability, not human intuition (Alexandr Wang): Things easy for humans (e.g., folding laundry) are extremely hard for AI due to a lack of digital data; things extremely hard for humans (e.g., proving mathematical theorems) can be easy for AI because there is abundant, verifiable digital data.

3. AI will first automate "white-collar" jobs, not "blue-collar" jobs (Alexandr Wang): Knowledge work involving Word and Excel all day, being entirely digitized, is more easily replaced by AI than physical labor in the real world.

4. The "power law effect" is stronger for AI than for software (Alexandr Wang): Top AI talent and systems (e.g., OpenAI's 250-person team) can have an outsized impact far beyond that of software companies, making high-skilled immigration policy a key national strategy.

5. AI's S-curve is far from its peak; its potential business value is 10-100 times that of software (Alexandr Wang): The software S-curve is saturating, while AI's S-curve is just beginning. AI can automate large-scale repetitive tasks within enterprises, creating business value far exceeding that of deploying CRM or ERP systems.

6. "Caring" is revealed in the details and the edges (Alexandr Wang): To judge if someone "gives a shit," don't look at their positive performance, but observe the depth, obsession, and energy they invest in details and edges that others overlook.

7. AI will lead to "problems being more valuable than answers" (Alexandr Wang): As tools like GitHub Copilot make coding itself easy, the differentiating human skill will shift from "how to build" to "what to build" — the ability to imagine and define the problem becomes core.