The Grammarly–Superhuman Megadeal, TWiST 500 chats with LabelBox and Apptronik’s founders | E2146

Key Concepts

• AI Regulation: State vs. Federal control over AI development.
• Pay Per Crawl: A model where websites charge AI companies for scraping their content.
• RAG (Retrieval Augmented Generation): AI models collecting information from the internet to generate responses.
• AI Native Productivity Suite: A collection of AI-powered tools and agents designed to enhance productivity.
• AI Agents: AI systems that work on behalf of users to automate tasks and provide assistance.
• Data Labeling: The process of annotating data to train AI models.
• Data Factory: A verticalized stack combining a skilled workforce, technology platform, and AI to produce high-quality training data.
• Humanoid Robotics: The development and application of robots with human-like form and capabilities.
• Task Mastery: Achieving human-level performance in a specific task by a robot.
• AI Superhighway: A platform that brings AI to users across various applications and devices.

AI Regulation: State vs. Federal Control

• Issue: A proposal to limit state-level AI regulation for 10 years was considered but appears to be off the table for now.
• Arguments:
  • Pro-Federal Regulation: AI companies prefer a single national policy to avoid the complexity of 50 different state regulations.
  • Anti-Federal Regulation: States should have some power to regulate AI, with common-sense limits to avoid hindering the entire American AI industry.
• Perspective: A federal-level rule might be beneficial, but a complete ban on state regulation is too aggressive.

Cloudflare's Pay Per Crawl: Revolutionizing Online Content Economics

• New Product: Cloudflare introduced "Pay Per Crawl," allowing websites to block AI scraping or set a price for it.
• Context:
  • Training: AI models scrape the internet to train their systems.
  • RAG: AI models collect information to generate responses for users.
• Problem: RAG queries don't send traffic back to the original websites, creating a one-sided value transfer.
• Solution: Pay Per Crawl enables websites to charge for scraping, with Cloudflare mediating transactions.
• Potential Impacts:
  • Better online media economics.
  • More expensive AI subscriptions.
  • Stronger online content ecosystem.
  • Better and stronger AI systems in the long term.
• Challenges:
  • Uptake: How many websites will adopt this model?
  • Compliance: Will AI companies agree to pay?
• Alternative Solutions: AI outputs could be structured to send users to third-party sources, similar to Google search results.

Grammarly Acquires Superhum: Building an AI Native Productivity Suite

• Acquisition: Grammarly acquired Superhum to expand its AI-powered productivity suite.
• Vision: Grammarly aims to build a platform of AI agents that work seamlessly within existing tools.
• Rationale:
  • Email is a dominant communication tool and Grammarly's number one use case.
  • Superhum has reinvented email and offers a mind shift in how people think about it.
• Integration:
  • Agents will follow users across different applications.
  • KOD and Superhum will serve as first-party homes for interacting with those agents.
• Multi-Agent Framework: Users can have multiple agents working simultaneously, such as sales, support, and marketing agents.
• Agent Marketplace: Grammarly plans to create a marketplace where others can build and deploy agents on its platform.
• Superhum's Future: Superhum will retain its identity and accelerate its development with Grammarly's resources.
• Email as a Workbench: Email can become a central place to manage agents and tasks.
• Grammarly's AI Superhighway: A platform that brings AI to users across various applications and devices.
• Assurances for Superhum Users: The Superhum experience will not change, and the team will continue to invest in its core features.

Labelbox: AI Data Labeling and the Data Factory Concept

• Data Labeling: Annotating data to train AI models, now often AI-assisted.
• AI-Assisted Human Data Labeling: Experts use AI systems to produce high-quality data for training models.
• Data Factory: A verticalized stack that includes:
  • A network of high-end talent with superior knowledge.
  • A technology platform with tools, workflows, and AI.
• Pricing Model: Labelbox prices its data services based on the value the data brings to customers.
• Market Demand: Demand for Labelbox's services has increased substantially, especially with the rise of GenAI.
• Potential Revenue: The sale of Scale AI to Meta could generate hundreds of millions in new revenue for Labelbox.
• M&A Activity: Labelbox is seeing increased M&A interest in the data labeling space.
• Interacting with AI Models: Data is the primary way to communicate with and manage AI models.

Apptronic: Humanoid Robots and the Future of Automation

• Apollo Robot: Apptronic's humanoid robot, designed for various tasks.
• Apollo 2: A more robust and scalable version of Apollo, currently in the pilot stage.
• Robustness: A stable hardware platform that can be used for training and building models.
• Learning-Based Robotics: The future of robotics lies in learning, not hard coding.
• Task Mastery: Achieving human-level performance in specific tasks.
• ROI Considerations: The cost of the robot relative to the cost of human labor, considering factors like labor shortages and fully burdened labor rates.
• Google DeepMind Partnership: Collaboration to push the boundaries of what's possible in humanoid robotics.
• China's Competitiveness: China is a real contender in the humanoid robotics space, with a national strategy and significant funding.
• National Robotics Strategy: The US needs a national robotics strategy to compete with China.
• Incentivizing Uptake: The US should incentivize businesses to adopt American robots.
• Humanoid Robotics as the PC Era: Humanoid robots are at the beginning of a 40-year cycle, similar to the PC era.
• Austin, Texas as a Hub: Austin is emerging as a hub for hardware companies, with access to the Texas-Mexico corridor for manufacturing.

Synthesis/Conclusion

The transcript highlights several key trends in the tech industry: the increasing importance of AI regulation, the evolving economics of online content, the rise of AI-powered productivity tools, the growing demand for high-quality training data, and the advancements in humanoid robotics. The discussions around Pay Per Crawl, Grammarly's acquisition of Superhum, Labelbox's data factory concept, and Apptronic's Apollo robot provide valuable insights into the challenges and opportunities in these rapidly developing fields. The need for collaboration, strategic investment, and a focus on real-world applications are emphasized as crucial factors for success in the AI and robotics industries.