David Rein

Email Google Scholar GitHub Twitter

About

I'm a researcher at METR, trying to shine a flashlight to help us navigate the cliff's edge we're walking along with AI.

I think there's a very good chance that humanity is on track to create AI systems substantially smarter and more capable than us. We need to be really careful about this, because it's an open scientific question whether we can control such systems.

I'm most well-known for creating GPQA, a widely used AI capability benchmark. I've also done research on scalable oversight (i.e. trying to figure out how to supervise a system that's smarter or more capable than the supervisor), and I was one of the first few employees at Cohere, where I trained embedding models for semantic search.

Research highlights

In the news

Writing & talks

Essays 7
  • A New Color

    2026 · A paper that produced a genuinely new color—an impossibly saturated teal—by laser-stimulating individual photoreceptors in the eye.

  • AI Companies as 'Hot Springs' for Growing AIs

    2026 · The "rogue AI escapes containment" framing misses that AI labs are actually the most attractive place for a misaligned AI to stay.

  • The Compute Bottleneck Intuition Is Formed Under Unusual Conditions

    2026 · AI researchers are biased against fast takeoff scenarios because our intelligence is largely fixed.

  • WTBU

    2026 · A small communication technology, borrowed from nuclear submarines, that lets you flag possibly-obvious things to colleagues without ego getting in the way.

  • A Message to My Friends About AI

    2026 · A message I sent to close friends about how my view on AI shifted, and why I now expect the coming transition to be radically chaotic.

  • Science Comms at METR

    2025 · Some notes on how we communicate nuanced results at METR while preserving clarity.

  • Can Good Benchmarks Contain Mistakes?

    2024 · Why benchmarks can still be useful even when they aren't perfect, and how I think about their limits.

Research 5
Talks & podcasts 3
David Rein