Epoch AI's Greg Burnham on AI cracking decades-old math problems like Navier-Stokes
Jsevillamol · x · 2026-09-30
TWIML AI podcast hosts Greg Burnham, who leads AI capabilities research at Epoch AI, on how AI went from failing grade-school math to helping solve research problems that resisted mathematicians for decades, including Navier-Stokes.
Key points:
- How much progress relies on persistence and prior human work vs. genuinely new ideas.
- Measuring progress as traditional benchmarks lose usefulness.
- Why capability gains appear surprisingly steady across model generations.
- Where models still struggle: open-ended work, learning from experience, spotting promising research directions.
More from AGI Musings
- One person with chained models now delivers music videos studios used to staff teams for — minchoi · 2026-09-30
- Peter Diamandis: civilizations are constrained by power, more energy means more possibility — PeterDiamandis · 2026-09-30
- Ben Recht Critiques NFL Win-Probability Models Claiming Three-Digit Precision — beenwrekt · 2026-09-30
- Founders Can Build Around a Hard Line, Not a Maybe: Sanders AI Plan Chilling Effect — victor_explore · 2026-09-30
- Developer builds AI relationship coach Agent Eros and invites Reddit to argue against it — Spiritual-Fold6038 · 2026-09-30
- A Mathematician's Ground Truth: AI Has Barely Changed Day-to-Day Math Research — littmath · 2026-09-30