Skip to content
No PriorsNo Priors

No Priors Ep. 65 | With Scale AI CEO Alexandr Wang

Alexandr Wang was 19 when he realized that gathering data will be crucial as AI becomes more prevalent, so he dropped out of MIT and started Scale AI. This week on No Priors, Alexandr joins Sarah and Elad to discuss how Scale is providing infrastructure and building a robust data foundry that is crucial to the future of AI. While the company started working with autonomous vehicles, they’ve expanded by partnering with research labs and even the U.S. government. In this episode, they get into the importance of data quality in building trust in AI systems and a possible future where we can build better self-improvement loops, AI in the enterprise, and where human and AI intelligence will work together to produce better outcomes. Sign up for new podcasts every week. Email feedback to show@no-priors.com Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @alexandr_wang 0:00 Introduction 3:01 Data infrastructure for autonomous vehicles 5:51 Data abundance and organization 12:06 Data quality and collection 15:34 The role of human expertise 20:18 Building trust in AI systems 23:28 Evaluating AI models 29:59 AI and government contracts 32:21 Multi-modality and scaling challenges

Sarah GuohostAlexandr (Alex) WangguestElad Gilhost
May 21, 202439mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Scale AI’s Alexandr Wang on data abundance, evals, and AGI’s path

  1. Alexandr Wang, CEO of Scale AI, explains how Scale evolved from powering autonomous vehicle datasets to becoming the core “data foundry” behind nearly every major large language model and key government AI programs.
  2. He argues that AI’s limiting factor is shifting from compute to high-quality, expert-driven data, and outlines a vision of “data abundance” built from proprietary corpora, expert annotations, and hybrid human–AI synthetic data.
  3. Wang emphasizes the importance of rigorous evaluations and public leaderboards to properly measure model capabilities, build trust, and support safe deployment across enterprises, governments, and consumer applications.
  4. He believes the path to AGI will be gradual and domain-by-domain—more like curing cancer than inventing a single vaccine—with humans remaining crucial partners in guiding, critiquing, and extending AI systems over long time horizons.

IDEAS WORTH REMEMBERING

5 ideas

Data is becoming the primary bottleneck for AI progress.

While compute spending is measured in tens or hundreds of billions, Wang argues that moving from GPT‑4 to GPT‑10 will be constrained by the availability of diverse, high-quality data rather than just more GPUs.

High-quality ‘frontier data’ matters far more than raw volume.

Enterprise and internet-scale datasets are huge, but only a small, carefully filtered subset—expert reasoning traces, agent workflows, multilingual and multimodal data—actually drives meaningful model improvements.

Hybrid human–AI pipelines will define the future of data generation.

Models can generate large amounts of initial content, but human experts are needed to correct, critique, and refine outputs to produce reliable synthetic data that meaningfully upgrades model capabilities.

Robust, held-out evaluations are essential to trust and safety.

Existing public benchmarks are often in training data and overfit; Scale is building private, regularly refreshed evals and leaderboards so labs, governments, and enterprises can accurately understand model strengths and weaknesses.

Every serious AI application will need a self-improvement loop.

Wang notes that leading labs succeed by continuously collecting usage data and evals to refine models; he expects enterprises and governments will need similar data flywheels, which Scale’s Gen AI platform aims to enable.

WORDS WORTH SAVING

5 quotes

AI in general is the product of three fundamental pillars: the algorithms, the compute, and the data.

— Alexandr Wang

We as an industry can either choose data abundance or data scarcity, and we view our role to be to build data abundance.

— Alexandr Wang

Producing high-quality data for AI systems is near infinite impact, because even a tiny improvement in a model compounds over every future invocation.

— Alexandr Wang

The question is not whether a model is better than a human; the question is whether a human plus a model is better than a model alone.

— Alexandr Wang

The path to AGI looks a lot more like curing cancer than developing a vaccine.

— Alexandr Wang

Founding story and evolution of Scale AI from AV data to LLMsThe three pillars of modern AI: compute, algorithms, and dataData abundance vs. data scarcity and the importance of frontier dataHybrid human–AI data generation and expert-driven supervisionAI evaluation challenges, benchmarks, and public leaderboardsEnterprise and government AI applications and self-improving systemsLong-term outlook on AGI, human–AI complementarity, and industry trajectory

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome