Scott Aaronson: AI one-shot open problems from my 25-year-old papers
jon_barron · x · 2026-10-17
Former OpenAI researcher and UT Austin professor Scott Aaronson says he fed several open problems from his 25-year-old papers to AI this summer, and the models solved them all one-shot. He estimates each would have taken him at least 100 hours of focused work — time he no longer has in contiguous blocks.
He also admits this convenience is accelerating the atrophy of his own math skills: for anything solvable in a few hours that GPT can answer instantly, he often just lets the model do it.
More from AGI Musings
- Billions of AI agents are coming: why every machine needs a digital ID passport — mikeflache · 2026-10-17
- Kenyan founder built AI platform for $25K with Chinese open-source model instead of $1M Claude bill — 96Stats · 2026-10-17
- Rachel Thomas: 'Perfect' AI agents are neither realistic nor desirable, human oversight is here to stay — math_rachel · 2026-10-17
- "BMW blinker installers still get paid": a one-liner on why AI job fears are overblown — justalexoki · 2026-10-17
- In the AI Era, Execution Is Cheap — Ideas Are What's Scarce Now — djcows · 2026-10-17
- Security Researcher Mocks AI Hype Amid Rising Tech Unemployment — evilsocket · 2026-10-17