20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data...

Full Title

20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Summary

This episode explores the evolving landscape of AI and data labeling, with Jonathan Siddharth of Turing arguing that data labeling companies are becoming obsolete, replaced by "research accelerators." The discussion delves into the increasing complexity of data required for advanced AI models, the shift from chatbots to agentic AI, and the future of knowledge work, suggesting that a significant portion will be automated, leading to profound societal and economic changes.

Key Points

• The data labeling market is evolving beyond simple tasks to complex, domain-specific data generation, requiring expert humans for advanced AI model training.
• The shift from training AI for tests to training for real-world economically valuable work necessitates more sophisticated data and agentic capabilities.
• Turing positions itself as a "research accelerator," not a talent marketplace, focusing on training superintelligence for frontier AI labs by providing compute, research, and complex data.
• The development of AI agents requires specialized training data generated through reinforcement learning environments, simulating real-world business workflows.
• Enterprises will increasingly require custom-tuned, often smaller, AI models for specific workflows due to data privacy concerns and the need for efficiency, a permanent requirement rather than a temporary phase.
• The automation of knowledge work is inevitable, driven by increasingly capable AI agents, which will lead to a significant transfer of budget from human labor to AI technology.
• While AI adoption in back-office functions may be slow, front-office applications, particularly in finance and life sciences, are expected to see faster integration due to direct revenue-generating potential.
• The core moat in the AI era is shifting from technology to data feedback loops, where continuous learning from real-world deployment enhances model capabilities.
• "First-mile" and "last-mile" challenges, involving data acquisition, structuring, and workflow integration, remain significant obstacles in AI deployment.
• The future of AI will involve intelligent agents that can execute complex, multi-step workflows, potentially reducing the need for traditional SaaS applications as AI becomes more integrated.
• The rapid pace of AI development and the high stakes of winning the "superintelligence race" justify significant investment, even with potential market volatility.
• The definition of a software engineer will broaden, enabling more non-technical individuals to build software products with the aid of AI tools, leading to an expansion of software creation.
• The future interface for AI may move beyond smartphones to more integrated, always-on devices that process multimodal data and act as extensions of human cognition.
• The data provisioning market will reward companies with strong research DNA and the ability to adapt quickly to the rapidly changing AI landscape, leading to a few dominant players.
• While AI capabilities are rapidly advancing, a slow and steady takeoff is more likely, providing humanity time to adapt the workforce and education systems.

Conclusion

The data labeling market is fundamentally changing, moving towards more complex, expert-driven data generation for advanced AI models.

The rise of agentic AI necessitates a shift in how models are trained and deployed, requiring specialized data and environments to simulate real-world workflows.

The future promises widespread automation of knowledge work, leading to increased productivity and a redefinition of human roles and the job market.

Discussion Topics

• How do you see the role of human expertise evolving in AI development as models become more sophisticated?
• What are the biggest ethical considerations we need to address as AI increasingly automates knowledge work?
• Beyond productivity, what are the most significant societal benefits or challenges you anticipate from widespread AI adoption in the next decade?

Key Terms

GMV

Gross Merchandise Volume - the total value of merchandise sold over a given period.

SaaS

Software as a Service - a software licensing and delivery model where software is licensed on a subscription basis and is centrally hosted.

AGI

Artificial General Intelligence - AI with the capacity to understand or learn any intellectual task that a human being can.

LLM

Large Language Model - a type of AI model trained on a massive amount of text data to understand and generate human-like text.

SFT

Supervised Fine-Tuning - a method of training AI models by providing them with examples of desired input-output pairs.

RLHF

Reinforcement Learning from Human Feedback - a technique used to align AI models with human preferences by using human feedback to train a reward model.

Reinforcement Learning

A type of machine learning where an agent learns to make decisions by taking actions in an environment to maximize a reward.

Agentic AI

AI systems designed to act autonomously and take actions in the real world or digital environments to achieve specific goals.

On-prem

On-premises - software that is installed and runs on computers on the premises of the person or organization that uses the software, rather than at a remote service provider's location.

FDE

Field Data Engineer - a role that likely involves on-site or direct interaction with data and systems to implement AI solutions.

Alpha

In finance, alpha represents the excess return of an investment relative to the return of a benchmark index.

PageRank

An algorithm used by Google Search to rank web pages in their search engine results.

ClickStream

Data generated by users' interactions with a website, such as page views and clicks.

S&P 500

A stock market index that represents the stocks of 500 large-cap U.S. companies.

GDPVal

Likely refers to a specific paper or metric related to AI's economic value or impact on GDP.

NLP

Natural Language Processing - a branch of AI that deals with the interaction between computers and human language.

MVP

Minimum Viable Product - the version of a new product which allows a team to collect the maximum amount of validated learning about customers with the least effort.

Pos System

Point of Sale System - the system where a customer pays for goods or services.

CRM

Customer Relationship Management - a strategy and technology for managing all your company's relationships and interactions with customers and potential customers.

Timeline