Interview, Fireside Chat
Turing CEO Jonathan Siddharth: Who Wins in Data Labelling & Why 99% of Knowledge Work Will Disappear
- Turing has pivoted from a talent marketplace to a "research accelerator" and is now training superintelligence for seven of the eight frontier AI labs.
- Jonathan Siddharth asserts the era of traditional data labeling companies is over, replaced by the need for complex, real-world data to train agentic systems.
- The data generation paradigm has shifted from simple tasks (e.g., sorting numbers) to complex, multi-step workflows (e.g., building a full B2B marketplace app).
- A key distinction in the new data requirements is the move from teaching AI to pass tests to teaching AI to perform economically valuable work like a lawyer or paralegal.
- The industry has transitioned from chatbots (SFT/RLHF) to "agents" capable of executing multi-step real-world workflows using tool use and API calls.
- Turing is building massive Reinforcement Learning (RL) environments to train agents, creating a "mini world model" for every workflow, role, and industry function.
- Siddharth estimates that approximately $30 trillion worth of digital knowledge work could be automated if AI agents reach full competence.
- The company operates in "Innings 1" of vertical data acquisition, believing the market for specialized workflow data is vast and not yet saturated.
- Turing differentiates itself by combining deep research DNA with real-world enterprise deployment, serving both frontier labs and clients like Disney, Pepsi, and BlackRock.
- The firm provides Fine-tuned Deployments (FDEs) for enterprises to build smaller, on-prem models (500M to 10B parameters) to protect proprietary data and reduce latency.
- Siddharth predicts a 10-to-20-year decline for incumbents that fail to adopt new AI tools, as startups will outcompete them by operating with significantly lower head counts.
- He observes that front-office automation (e.g., investment decisions) is advancing faster than back-office automation due to direct revenue incentives and market efficiency pressures.
- Citing an OpenAI paper on GDP, Siddharth notes current models achieve parity with human experts in roughly 50% of specific, real-world tasks.
- The speaker believes the future of entrepreneurship will explode as founders can recruit "GPTs" for software engineering and marketing, reducing the capital needed to launch.
- He rejects the idea that AI will lead to widespread idleness, arguing humans will shift to solving problems at higher levels of abstraction, such as curing diseases.
- Data-driven feedback loops, where model deployments generate new data to improve subsequent iterations, are identified as the primary future moat for AI applications.
- The "first mile schlep" involves enterprises structuring fragmented, messy data, while the "last mile schlep" involves designing workflows for partial autonomy and human-AI collaboration.
- Turing is shifting pricing from time-based billing to value-oriented models as the technology matures.
- Siddharth does not see an AI bubble, citing a significant "model capability overhang" where current models are underutilized without proper agentic scaffolding.
- He predicts the end of traditional SaaS as we know it, as companies will build custom vertical solutions or rely on agentic foundation models rather than maintaining 80+ generic SaaS tools.
- The primary interface of the future is envisioned as a multimodal wearable device (e.g., glasses or earbuds) that acts as an "extension of the brain" rather than a phone.
- The data provisioning market is expected to consolidate among a "few winners" who possess strong research DNA and the agility to adapt to rapidly changing learning paradigms.
- Siddharth believes the market for robotics and embodied AI remains a vast, early-stage opportunity compared to the more crowded digital intelligence sector.
- He advocates for a "slow takeoff" of AGI to allow humanity time to upskill and adjust the workforce, rather than facing a sudden, disruptive shift.
- Turing is transitioning from a fully distributed team to a hub-and-spoke model with offices in San Francisco, Palo Alto, and London to increase collaboration.
- Siddharth's personal leadership philosophy has shifted from high-leverage detachment to deep, hands-on involvement with customers and engineering teams.
- His greatest excitement for the next decade is using AI to automate research discovery, specifically aiming for breakthroughs in MS drug discovery.