Inference by Turing Post What Really Blocks AI Progress? Ulrik Hansen from Encord thinks it’s…
Sep 25, 2025
Ulrik Hansen, co-founder of Encord, shares insights on the real roadblocks to AI progress. He argues that data, not models, is the true bottleneck, emphasizing the importance of data orchestration. He contrasts Tesla's live feedback system with Waymo's cautious rollout, discussing the challenges in robotics and edge cases in self-driving tech. Ulrik highlights the shift towards a connection economy and the rising value of trust in brands. He also explores the potential pitfalls of synthetic data and why applied AI is more thrilling than abstract AGI discussions.
AI Snips
Chapters
Books
Transcript
Episode notes
Embodiment Makes Feedback Harder
- Robotics shares self-driving's data challenges but adds hardware, latency, and connectivity constraints.
- Physical embodiment makes building fast feedback loops and on-edge models substantially harder.
Engineer Specialized Data Tooling
- Build custom tooling per use case for human feedback and data orchestration.
- Treat AI toolchains like software ecosystems with versioning, telemetry, and protocols to scale safely.
Humans Provide Preference And Edge Cases
- Humans will supply preferences and edge-case judgments that models cannot infer.
- In enterprise workflows, experts provide ongoing signals as they do their daily work, shaping model behavior.
