# Greg Burnham

Markdown copy of https://realalpha.app/people/greg-burnham/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

Real Alpha has cut 8 moments of Greg Burnham (Lead of AI Capabilities Research, Epoch AI) talking, from an interview on The TWIML AI Podcast published Sep 29, 2026.

- Role: Lead of AI Capabilities Research, Epoch AI

## What Greg Burnham said

- [AI math has advanced from school problems to open research](https://realalpha.app/s/ai-math-has-advanced-from-school-problems-to-open-research/)
  The TWIML AI Podcast · Sep 29, 2026 · 7 min
  > “AI can solve problems that humans have cared about just independently.” — Greg Burnham · [25:45](https://realalpha.app/v/lz9J-Dbj2zw/?t=1545)
- [AI uses accumulated knowledge and persistence in hard math](https://realalpha.app/s/ai-uses-accumulated-knowledge-and-persistence-in-hard-math/)
  The TWIML AI Podcast · Sep 29, 2026 · 6 min
  > “One is it knows everything. It knows all the prior work. It's got the literature memorized.” — Greg Burnham · [31:03](https://realalpha.app/v/lz9J-Dbj2zw/?t=1863)
- [Benchmark gains mix targeted training and generalization](https://realalpha.app/s/benchmark-gains-mix-targeted-training-and-generalization/)
  The TWIML AI Podcast · Sep 29, 2026 · 7 min
  > “that deep generalization means you could really see, if that happened in a rich way where you just didn't have to even be trained on something at all to gain capability in that area” — Greg Burnham · [11:39](https://realalpha.app/v/lz9J-Dbj2zw/?t=699)
- [Compute and data underpin most measured AI progress](https://realalpha.app/s/compute-and-data-underpin-most-measured-ai-progress/)
  The TWIML AI Podcast · Sep 29, 2026 · 5 min
  > “until you see the data and compute supply chain ramp significantly or discontinuously, to be more precise, you're not likely to see discontinuities on the model side.” — Greg Burnham · [15:29](https://realalpha.app/v/lz9J-Dbj2zw/?t=929)
- [Experts judge novelty by how much of a hint humans needed](https://realalpha.app/s/experts-judge-novelty-by-how-much-of-a-hint-humans-needed/)
  The TWIML AI Podcast · Sep 29, 2026 · 8 min
  > “How big a hint would a human have needed to solve this in hindsight?” — Greg Burnham · [39:34](https://realalpha.app/v/lz9J-Dbj2zw/?t=2374)
- [Current harnesses do not make models learn strategy on their own](https://realalpha.app/s/current-harnesses-do-not-make-models-learn-strategy-on-their/)
  The TWIML AI Podcast · Sep 29, 2026 · 5 min
  > “we've tried pretty hard and have mostly not found anything that cracks this” — Greg Burnham · [56:16](https://realalpha.app/v/lz9J-Dbj2zw/?t=3376)
- [Models still struggle with open-ended work and learning by practice](https://realalpha.app/s/models-still-struggle-with-open-ended-work-and-learning-by/)
  The TWIML AI Podcast · Sep 29, 2026 · 5 min
  > “gpt 5.6 scores x out of the box and then stays flat on its playthroughs” — Greg Burnham · [53:36](https://realalpha.app/v/lz9J-Dbj2zw/?t=3216)
- [Automatic checks make unsolved math problems usable as benchmarks](https://realalpha.app/s/automatic-checks-make-unsolved-math-problems-usable-as/)
  The TWIML AI Podcast · Sep 29, 2026 · 5 min
  > “This is maybe the big piece of work on our side, was to make it easy to verify automatically whether an AI system has solved this problem.” — Greg Burnham · [45:35](https://realalpha.app/v/lz9J-Dbj2zw/?t=2735)

## The episode

- [AI Progress Explained: From Math Olympiads to Navier-Stokes](https://realalpha.app/v/lz9J-Dbj2zw/) · The TWIML AI Podcast · Sep 29, 2026