20VC: Scale's Alex Wang on Why Data Not Compute is the Bottleneck to Foundation Model Performance, Why AI is the Greatest Military Asset Ever, Is China Really Two Years Behind the US in AI and Why the CCPs Industrial Approach is Better than Anyone Else's
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The PitchTue Jun 11 2024
Scale AI's Financial Performance and Growth:
- Scale AI has raised $1.6 billion for the company, achieving a reported valuation of $14 billion earlier this year.
- The company significantly increased its Annual Recurring Revenue (ARR) in 2023 and is projected to reach $1.4 billion in ARR by the end of 2024.
Scale AI has experienced substantial financial growth with significant funding and revenue projections, positioning itself as a major player in the industry.
AI as a Military Asset and CCP's Industrial Approach:
- Alex Wang highlighted that AI technology could potentially be a more significant military asset than nuclear weapons.
- China's centralized industrial policy under the CCP is praised for effectively driving forward critical industries through aggressive actions.
The discussion emphasized the potential of AI technology as a powerful military asset surpassing even nuclear capabilities. It also acknowledged China's successful industrial strategy under the CCP, showcasing effective centralized actions to advance key industries.
Data Production Strategies and Overcoming Data Scarcity:
- To advance AI capabilities beyond emulating internet data, frontier data containing complex reasoning chains, agent behavior, tool use, etc., is crucial.
- Two main strategies discussed are mining existing enterprise data and forward production of new frontier data through human-synthetic collaboration.
- Increasing the supply side of data involves longitudinal collection methods like process mining in workplaces or consumer devices capturing personal life activities.
The conversation delved into strategies essential for enhancing AI models beyond internet-based training by focusing on generating frontier data involving intricate reasoning processes. It outlined approaches such as leveraging enterprise data resources and creating new datasets collaboratively between humans and synthetic systems.
Value Capture Shifts and Customization Trends in Software:
- Value capture may shift from models themselves towards infrastructure below them (e.g., NVIDIA) and applications/services built on top (e.g., Salesforce).
- Enterprises are likely to demand greater customization leading to purpose-built software solutions tailored precisely to their needs.
Observations were made regarding potential shifts in value capture within the software industry towards infrastructure providers like NVIDIA and application developers such as Salesforce. There was an anticipation of heightened demand for customized software solutions meeting specific enterprise requirements.
Transition from Per-seat Pricing to Consumption-based Models:
- With increasing reliance on AI agents over human labor, per-seat pricing becomes less viable compared to consumption-based pricing models reflecting value provided by both people and AI systems.
The transition away from per-seat pricing towards consumption-based models was discussed due to evolving reliance on AI systems over human input, necessitating pricing structures aligning with contributions from both individuals and automated agents.
Regulatory Challenges Around Data Access and Innovation:
- Concerns exist about regulatory provisions stifling innovation due to Consumer Data Protection Acts and restrictive EU approach towards data access regulations.
- Balancing liberal democracy principles with ensuring efficient chip production-like efforts for maintaining high levels of manufacturing chips should also apply to managing data access regulations effectively.
Regulatory challenges surrounding innovation due to stringent consumer protection acts were highlighted along with concerns about hindrances posed by restrictive EU policies on accessing data. The importance of balancing democratic values with facilitating efficient chip production-like initiatives for robust management of data access rules was underscored.
Scale's Alex Wang on Foundation Models and Data Bottlenecks:
- Large data sets that do not provide proprietary advantages should be centralized for whole industries, like safety data in aerospace or fraud and compliance data in financial services.
- The need to work through existing restrictions to ensure AI progress, such as HIPAA regulations hindering the use of patient data for training AI models.
- Concerns about bold promises being divorced from technical reality in AI development, potentially leading to a hangover effect similar to what happened with autonomous vehicles.
China's Industrial Policy and AI Progress:
- China's approach to industrial policy is highlighted as effective due to its ability to drive critical industries forward aggressively.
- The rise of Chinese EV manufacturers showcases the success of China's industrial policies.
- Observations on China catching up in the AI race with advanced models like Yi-Large competing globally.
AI as a Military Asset and Closed Systems Debate:
- Discussion on how AI has the potential to be a powerful military asset, possibly more impactful than nuclear weapons.
- Consideration of closed systems for advanced AI technologies due to geopolitical concerns and security risks.
- Balancing open systems for economic value while keeping advanced systems closed for strategic advantage.
Media Treatment vs. Congressional Testimony:
- Alex Wang's perspective on media treatment versus congressional testimony regarding company PR and fair treatment.
- Importance of owning distribution channels and controlling the narrative directly rather than relying solely on traditional media outlets.
Company Building Principles and Hiring Practices at Scale:
- Importance of hiring individuals who deeply care about their work product and organization, willing to go the extra mile.
- Strategy at Scale focusing on maintaining high talent density by approving every hire personally to uphold an exceptionally high bar.
- Challenges faced during rapid team growth leading to difficulties in maintaining excellence within the organization.
Hypergrowth Strategies and Talent Ecosystems:
- Reflections on hypergrowth strategies impacting quality maintenance within teams over time.
- Importance of developing a self-preserving talent ecosystem independent of brand heat cycles for sustained excellence.