The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch 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
419 snips
Dec 1, 2025 
Jonathan Siddharth, Founder and CEO of Turing, dives into the future of AI and data labeling. He predicts that 99% of knowledge work will be automated within a decade, reshaping job markets. Siddharth discusses why traditional SaaS models are becoming obsolete in an AI-driven world and the critical need for custom, fine-tuned models in enterprises. He also highlights the importance of data-driven feedback loops as the new competitive moats, and shares insights into Turing's unique research-focused approach that sets it apart from legacy labeling firms.
AI Snips
Chapters
Transcript
Episode notes
Data Has Become Domain-Specific And Complex
- As models improved, data needs shifted from simple labels to complex, domain-specific datasets.
- Jonathan Siddharth says expert humans are now required to generate this higher-quality data.
Agents Need RL Environments, Not Just SFT
- Training agents requires environments, verifiers, and tool-use curricula rather than only SFT or RLHF.
- Siddharth explains RL environments let agents explore trajectories and create synthetic curricula for learning.
Create Realistic Simulated Workflows For Training
- Build RL environments that mimic real apps and instrument state so agents can practice workflows.
- Use synthetic databases and verifiers to reward successful multi-step tool use during training.